From 0a904aaaee299de113ff7d932800f8a1576e78ca Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?David=20Proch=C3=A1zka?= <david@prochazka.dev>
Date: Mon, 14 Dec 2020 10:48:06 +0100
Subject: [PATCH] DEL: similarityoperators source code
---
.../cover/AbstractHullRepresentation.java | 289 -------
.../cover/AbstractRepresentation.java | 809 ------------------
.../cover/BallRepresentation.java | 110 ---
.../similarityoperators/cover/CoverRank.java | 42 -
.../cover/FVDMRepresentation.java | 84 --
.../cover/HullBorderRepresentation.java | 142 ---
.../cover/HullCenterRepresentation.java | 45 -
.../cover/HullFastBorderRepresentation.java | 111 ---
.../cover/HullIncrementalRepresentation.java | 429 ----------
.../HullIncrementalRepresentationV2.java | 400 ---------
.../HullIncrementalRepresentationV3.java | 27 -
.../cover/HullKNHPRepresentation.java | 46 -
.../cover/HullMHPRepresentation.java | 43 -
.../cover/HullOptimizedRepresentation.java | 104 ---
.../cover/HullOptimizedRepresentationV2.java | 167 ----
.../cover/HullOptimizedRepresentationV3.java | 108 ---
.../cover/HullPreselectedRepresentation.java | 66 --
.../cover/HullRepresentation.java | 139 ---
.../cover/MinDistObjectPairIterator.java | 89 --
.../cover/SnakeAutoRepresentation.java | 382 ---------
.../cover/SnakeRepresentation.java | 134 ---
.../cover/types/BallBasedCover.java | 15 -
.../similarityoperators/groupBy/SGBAll.java | 63 --
.../similarityoperators/groupBy/SGBAll_N.java | 73 --
.../groupBy/SGBAll_N_Shuffle.java | 74 --
.../similarityoperators/groupBy/SGBAny.java | 46 -
.../similarityoperators/groupBy/SGBAny_N.java | 50 --
.../similarityoperators/groupBy/SGBQuery.java | 192 -----
.../similarityoperators/groupBy/SGBVote.java | 51 --
.../groupBy/SGBVote_N.java | 61 --
.../groupBy/SGBVote_N_Shuffle.java | 63 --
.../groupBy/SGB_All_Any.java | 6 -
.../groupBy/SoftSGBStrategies.java | 259 ------
.../operations/Containment.java | 22 -
.../operations/Intersection.java | 157 ----
.../operations/Subset.java | 23 -
.../similarityoperators/operations/Union.java | 152 ----
.../similarityoperators/utils/ImageNet.java | 89 --
.../utils/MergingType.java | 7 -
.../utils/ObjectGroups.java | 33 -
.../utils/OverlapType.java | 7 -
.../similarityoperators/utils/QueryType.java | 6 -
.../similarityoperators/utils/UnionFind.java | 126 ---
.../UnsupervisedBinaryClassification.java | 289 -------
src/messif/objects/keys/RadiusObjectKey.java | 186 ----
src/messif/utility/FileUtils.java | 38 -
...llRepresentationAsLocalAbstractObject.java | 107 ---
47 files changed, 5961 deletions(-)
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/AbstractHullRepresentation.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/AbstractRepresentation.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/BallRepresentation.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/CoverRank.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/FVDMRepresentation.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/HullBorderRepresentation.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/HullCenterRepresentation.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/HullFastBorderRepresentation.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/HullIncrementalRepresentation.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/HullIncrementalRepresentationV2.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/HullIncrementalRepresentationV3.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/HullKNHPRepresentation.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/HullMHPRepresentation.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/HullOptimizedRepresentation.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/HullOptimizedRepresentationV2.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/HullOptimizedRepresentationV3.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/HullPreselectedRepresentation.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/HullRepresentation.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/MinDistObjectPairIterator.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/SnakeAutoRepresentation.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/SnakeRepresentation.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/cover/types/BallBasedCover.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAll.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAll_N.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAll_N_Shuffle.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAny.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAny_N.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/groupBy/SGBQuery.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/groupBy/SGBVote.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/groupBy/SGBVote_N.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/groupBy/SGBVote_N_Shuffle.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/groupBy/SGB_All_Any.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/groupBy/SoftSGBStrategies.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/operations/Containment.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/operations/Intersection.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/operations/Subset.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/operations/Union.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/utils/ImageNet.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/utils/MergingType.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/utils/ObjectGroups.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/utils/OverlapType.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/utils/QueryType.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/utils/UnionFind.java
delete mode 100644 src/cz/muni/fi/disa/similarityoperators/utils/UnsupervisedBinaryClassification.java
delete mode 100644 src/messif/objects/keys/RadiusObjectKey.java
delete mode 100644 src/messif/utility/FileUtils.java
delete mode 100644 src/messif/utility/HullRepresentationAsLocalAbstractObject.java
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/AbstractHullRepresentation.java b/src/cz/muni/fi/disa/similarityoperators/cover/AbstractHullRepresentation.java
deleted file mode 100644
index 07c531e..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/AbstractHullRepresentation.java
+++ /dev/null
@@ -1,289 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import messif.objects.LocalAbstractObject;
-
-import java.io.IOException;
-import java.io.OutputStream;
-import java.util.ArrayList;
-import java.util.Arrays;
-import java.util.Iterator;
-import java.util.List;
-
-/**
- * A predecessor for all "hull-based" representations.
- * @param <T> type to return in {@link #build() }
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public abstract class AbstractHullRepresentation<T extends AbstractRepresentation> extends AbstractRepresentation<T> {
- /** List of hull objects */
- protected List<LocalAbstractObject> hull;
-
- public AbstractHullRepresentation(String locatorID, List<LocalAbstractObject> data) {
- super(locatorID, data);
- this.hull = new ArrayList<>();
- }
-
- public AbstractHullRepresentation(List<LocalAbstractObject> data) {
- this(null, data);
- }
-
- public AbstractHullRepresentation(PrecomputedDistances shared) {
- super(shared);
- this.hull = new ArrayList<>();
- }
-
- @Override
- public String toString() {
- String base = toShortString();
- StringBuilder sb = new StringBuilder(base.substring(0, base.length()-1));
- sb.append(", ");
- boolean sep = false;
- for (LocalAbstractObject c : hull) {
- if (sep)
- sb.append("; ");
- sb.append("center=").append(c);
- sep = true;
- }
-// for (LocalAbstractObject c : shared.data) {
-// if (sep)
-// sb.append("; ");
-// sb.append("base data=").append(c);
-// sep = true;
-// }
- sb.append("]");
- return sb.toString();
- }
-
-
- @Override
- public String toShortString() {
- StringBuilder sb = new StringBuilder(getClass().getSimpleName());
- sb.append('[');
- if (locatorID != null)
- sb.append("id=").append(locatorID).append(", ");
- sb.append("objects=").append(shared.getObjectCount()).append(", ");
- sb.append("centers=").append(hull.size()).append(", ");
- sb.append("coverage=").append(getCoverage()).append("]");
- return sb.toString();
- }
-
- @Override
- public void write(OutputStream out) throws IOException {
- for (LocalAbstractObject h : hull) {
- h.write(out);
- }
- }
-
- public void writeData(OutputStream out) throws IOException {
- for (LocalAbstractObject o : this.shared.data) {
- o.write(out);
- }
- }
-
- @Override
- public Iterator<LocalAbstractObject> getRepresentatives() {
- return hull.iterator();
- }
-
- /** @return List of hull objects */
- public List<LocalAbstractObject> getHull(){
- return hull;
- }
-
- public int[] getHullIdxs(){
- return shared.getIndexes(hull);
- }
-
- @Override
- public int getRepresentativesCount() {
- return hull.size();
- }
-
- protected void setCoveredByHull() {
- int[] hullIdxs = shared.getIndexes(hull);
- setCoveredByHullShared(hullIdxs);
-
- for (int i = 0; i < shared.getObjectCount(); i++) {
- if (isCovered(i)) // Skip hull objects that were set above
- continue;
- if (isCoveredByHull(i, hullIdxs.length, hullIdxs)) // checks for coverage by all hull points
- setCovered(i);
- }
- if (this.covered.cardinality() == this.shared.getObjectCount())
- allCovered = true;
- }
-
- protected void setCoveredByHullShared(int[] hullIdxs){
- clearCovered();
- if (hull.size() < 2)
- return;
-
- for (int idx : hullIdxs)
- setCovered(idx);
-
- }
-
- protected static boolean inHull(int idx, int[] idxs) {
- for (int i : idxs) {
- if (idx == i)
- return true;
- }
- return false;
- }
-
- /** This must be consistent with the computation in isCoveredByHull!!! */
- @Override
- public boolean isExternalCovered(LocalAbstractObject o) {
- return isExternalCovered(o, hull.size());
- }
-
- /** This must be consistent with the computation in isCoveredByHull!!! */
- public boolean isExternalCovered(LocalAbstractObject o, int closestHullObjsCnt) {
- int[] hullIdxs = shared.getIndexes(hull);
- return isExternalCovered(o, closestHullObjsCnt, hullIdxs);
- }
-
- /** This must be consistent with the computation in isCoveredByHull!!! */
- public boolean isExternalCovered(LocalAbstractObject o, int closestHullObjsCnt, int[] hullIdxs) {
- Ranked[] knns = getKNNRankedToExtObj(o, closestHullObjsCnt, hull, hullIdxs);
- if (knns[0].distance == 0) // A hull object itself matches
- return true;
- if (knns.length == 1) // Just one hull object, so only this hull object can match (which is handled above), so return false here.
- return false;
- // Must be consistent with isCoveredByHull! Here we circulate over all hull objects, not just the closest one!
- for (int i = 0; i < knns.length; i++) {
- if (!isExtObjCoveredByDistSum(knns, i, hullIdxs)) // hullIdxs[knns[i]], hullIdxs))
- return false;
- break; // Test just against the 1NN in hull objects!
- }
- return true;
- }
-
- public CoverRank getExternalCoverRank(LocalAbstractObject o) {
- int[] hullIdxs = shared.getIndexes(hull);
- Ranked[] knns = getKNNRankedToExtObj(o, hull.size(), hull, hullIdxs);
- CoverRank rank = null;
- // This must be in consistence with isCoveredByHull!!!
- for (int i = 0; i < knns.length; i++) {
- if (i == 0) {
- rank = getExtObjCoverRankByDistSum(knns, i, hullIdxs);
- if (!rank.covered)
- break;
- } else if (!isExtObjCoveredByDistSum(knns, i, hullIdxs)) { // hullIdxs[knns[i]], hullIdxs))
- rank.covered = false;
- break;
- }
- break; // Test just against the 1NN in hull objects!
- }
- return rank;
- }
-
- /**
- * @param knns array of k-nearest hull points (ranked instances that encapsulates the distance to the external object and index of the hull point in shared data)
- * @param nnIdx index to knns parameter that specifies which of them is the nearest one.
- * @param hullIdxs index of all hull objects in shared data
- * @return
- */
- private boolean isExtObjCoveredByDistSum(Ranked[] knns, int nnIdx, int[] hullIdxs) {
-// // Sum of distances to NNs (hull objects) but skip the selected NN (in nnIdx)
-// float sumObj = 0;
-// for (int i = 0; i < knns.length; i++) {
-// if (i != nnIdx)
-// sumObj += knns[i].distance;
-// }
-//
-// float[] distsNN = shared.getDistances(knns[nnIdx].index);
-// float sumNN = shared.sumArray(distsNN, hullIdxs);
-//
-// return (sumObj <= sumNN);
- float[] sums = computeDistanceSumsToNeighbors(knns, nnIdx, hullIdxs);
- return (sums[0] <= sums[1]);
- }
-
- /**
- * @param knns array of k-nearest hull points (ranked instances that encapsulates the distance to the external object and index of the hull point in shared data)
- * @param nnIdx index to knns parameter that specifies which of them is the nearest one.
- * @param hullIdxs index of all hull objects in shared data
- * @return
- */
- protected CoverRank getExtObjCoverRankByDistSum(Ranked[] knns, int nnIdx, int[] hullIdxs) {
-// // Sum of distances to NNs (hull objects) but skip the selected NN (in nnIdx)
-// float sumObj = 0;
-// for (int i = 0; i < knns.length; i++) {
-// if (i != nnIdx)
-// sumObj += knns[i].distance;
-// }
-//
-// float[] distsNN = shared.getDistances(knns[nnIdx].index);
-// float sumNN = shared.sumArray(distsNN, hullIdxs);
-// return new CoverRank((sumObj <= sumNN), (sumObj / sumNN) - 0.5f);
- float[] sums = computeDistanceSumsToNeighbors(knns, nnIdx, hullIdxs);
- final boolean isCovered = (sums[0] <= sums[1]);
- return new CoverRank(isCovered,
- (isCovered) ? (sums[0] / sums[1]) - 0.5f // [0;0.5]
- : (sums[0] / sums[1]) + 10f ); // (10;infty)
- }
-
- private float[] computeDistanceSumsToNeighbors(Ranked[] knns, int nnIdx, int[] hullIdxs) {
- // Sum of distances to NNs (hull objects) but skip the selected NN (in nnIdx)
- float sumObj = 0;
- for (int i = 0; i < knns.length; i++) {
- if (i != nnIdx)
- sumObj += knns[i].distance;
- }
-
- float[] distsNN = shared.getDistances(knns[nnIdx].index);
- float sumNN = shared.sumArray(distsNN, hullIdxs);
-
- return new float[] { sumObj, sumNN };
- }
-
- private Ranked[] getKNNRankedToExtObj(LocalAbstractObject o, int k, List<LocalAbstractObject> hull, int[] hullIdxs) {
- Ranked[] ranked = new Ranked[hull.size()];
- for (int i = 0; i < ranked.length; i++) {
- ranked[i] = new Ranked(o.getDistance(hull.get(i)), hullIdxs[i]);
- }
- Arrays.sort(ranked);
- return Arrays.copyOf(ranked, Math.min(k, ranked.length));
- }
-
- protected boolean isCoveredByHull(int objIdx, int closestHullObjsCnt, int[] hullIdxs) {
- int[] knns = getKNNIndexes(objIdx, closestHullObjsCnt, hullIdxs);
- // Must be consistent with isExternalCovered! Here we circulate over all hull objects, not just the closest one!
- for (int i = 0; i < knns.length; i++) {
- if (!isCoveredByDistSum(objIdx, hullIdxs[knns[i]], hullIdxs))
- return false;
- break; // Test just against the 1NN in hull objects!
- }
- return true;
- }
-
- protected boolean isCoveredByDistSum(int objIdx, int nnIdx, int[] nnIdxs) {
- float[] distsObj = shared.getDistances(objIdx);
- // Sum of distances to NNs but skip the selected NN (in nnIdx)
- float sumObj = shared.sumArray(distsObj, nnIdxs) - distsObj[nnIdx];
-
- float[] distsNN = shared.getDistances(nnIdx);
- float sumNN = shared.sumArray(distsNN, nnIdxs);
-
- return (sumObj <= sumNN);
- }
-
- /** Based on F-score, where importance is put on coverage than hull size. */
- public float getCoverageScore() {
- float beta = 1f; // or 2f
- float beta2 = beta * beta;
-
- float coverage = getCoverage(); // aka precision
- float hullSize = (float)(shared.getObjectCount() - hull.size()) / (float)(shared.getObjectCount() - 3); // aka recall
-
- return (1 + beta2) * coverage * hullSize / (beta2 * coverage + hullSize);
- }
-
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/AbstractRepresentation.java b/src/cz/muni/fi/disa/similarityoperators/cover/AbstractRepresentation.java
deleted file mode 100644
index 7d76d27..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/AbstractRepresentation.java
+++ /dev/null
@@ -1,809 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import messif.objects.LocalAbstractObject;
-import messif.statistics.Statistics;
-
-import java.io.IOException;
-import java.io.OutputStream;
-import java.text.DateFormat;
-import java.util.*;
-import java.util.concurrent.ExecutorService;
-import java.util.concurrent.Executors;
-import java.util.concurrent.TimeUnit;
-import java.util.logging.Level;
-import java.util.logging.Logger;
-
-/**
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- * @param <T> type to return in {@link #build() }
- */
-public abstract class AbstractRepresentation<T extends AbstractRepresentation> {
- /** ID of this representations, e.g. cluster ID, subset name */
- protected String locatorID;
-
- /** Data and precomputed distances */
- protected final PrecomputedDistances shared;
-
- /** Flags of being covered for each data object */
- protected final BitSet covered;
- /** Flag to signal that all objects are covered. */
- protected boolean allCovered;
-
- public AbstractRepresentation(String locatorID, List<LocalAbstractObject> data) {
- this.shared = new PrecomputedDistances(data);
- this.covered = new BitSet(data.size());
- this.locatorID = locatorID;
- }
-
- public AbstractRepresentation(List<LocalAbstractObject> data) {
- this(null, data);
- }
-
- public AbstractRepresentation(PrecomputedDistances shared) {
- this.shared = shared;
- this.covered = new BitSet(shared.getObjectCount());
- this.locatorID = null;
- }
-
- public final T build() {
- long timeStart = System.currentTimeMillis();
- System.err.println(timeStart + " started");
- select();
- long timeStop = System.currentTimeMillis();
- System.err.println("Selection by " + getClass().getSimpleName() + " took: " + (timeStop - timeStart) + " msec");
- return (T)this;
- }
-
- public String getLocatorID() {
- return locatorID;
- }
-
- public void setLocatorID(String locatorID) {
- this.locatorID = locatorID;
- }
-
- /**
- * Implementation of representation selection algorithm goes here.
- * This part is timed, so the duration in msec will be printed to stderr.
- */
- protected abstract void select();
-
- /** Method for checking whether an external object (out of passed list of objects) is covered by this representation or not.
- * @param o external object
- * @return decision whether the object is covered by this representation or not
- */
- public boolean isExternalCovered(LocalAbstractObject o) {
- return false;
- }
-
- protected final boolean isCovered(LocalAbstractObject o) {
- return isCovered(shared.getIndex(o));
- }
-
- protected final boolean isCovered(int idx) {
- return covered.get(idx);
- }
-
- protected final void setCovered(LocalAbstractObject o) {
- setCovered(shared.getIndex(o));
- }
-
- protected final void setCovered(int idx) {
- covered.set(idx);
- }
-
- public final List<LocalAbstractObject> getNotCovered(){
- List<LocalAbstractObject> notCovered = new ArrayList<>();
- for (int i = 0; i < shared.getObjectCount(); i++){
- if (!isCovered(i)) {
- notCovered.add(shared.getObject(i));
- }
- }
- return notCovered;
- }
-
- public void printNotCovered(OutputStream out) throws IOException {
- List<LocalAbstractObject> notCovered = getNotCovered();
- for (LocalAbstractObject object : notCovered) {
- object.write(out);
- }
- }
-
- protected final void clearCovered(LocalAbstractObject o) {
- clearCovered(shared.getIndex(o));
- }
-
- protected final void clearCovered(int idx) {
- covered.clear(idx);
- allCovered = false;
- }
-
- protected final void clearCovered() {
- covered.clear();
- allCovered = false;
- }
-
- protected final boolean isAllCovered() {
- if (allCovered)
- return true;
- allCovered = (covered.cardinality() == shared.getObjectCount());
- return allCovered;
- }
-
- /** Percentange of coverage of the data-set by the selected cover points.
- * @return value between 0 and 1 corresponding to coverage. The objects selected as cover points are not included in the computation.
- * I.e. coverage = |covered - cover_points| / |data_set - cover_points|
- */
- public final float getCoverage() {
- if (allCovered)
- return 1f;
- int excl = getRepresentativesCount();
- return (float)(covered.cardinality() - excl) / (float)(shared.getObjectCount() - excl);
- }
-
- protected void reportNonCovered() {
- int from = 0;
- boolean firstPrint = true;
- while ((from = this.covered.nextClearBit(from)) < this.covered.length()) {
- if (firstPrint) {
- System.err.println("# Reporting non-covered objects: " + DateFormat.getInstance().format(Calendar.getInstance().getTime()));
- firstPrint = false;
- }
- System.err.println("# Non-covered: " + this.shared.getObject(from).getLocatorURI());
- from++;
- }
- }
-
- protected void setCoveredByBall(LocalAbstractObject center, float radius) {
- int centerIdx = shared.getIndex(center);
- float[] dists = shared.getDistances(centerIdx);
- for (int i = 0; i < shared.getObjectCount(); i++) {
- if (dists[i] <= radius)
- setCovered(i);
- }
- }
-
- protected LocalAbstractObject getMedoid() {
- float min = Float.MAX_VALUE;
- LocalAbstractObject medoid = null;
-
- for (int i = 0; i < shared.getObjectCount(); i++) {
- float sum = shared.sumArray(shared.getDistances(i));
- if (sum < min) {
- min = sum;
- medoid = shared.getObject(i);
- }
- }
-
- return medoid;
- }
-
- protected LocalAbstractObject getMedoid(List<LocalAbstractObject> lst) {
- int[] idxs = shared.getIndexes(lst);
-
- float min = Float.MAX_VALUE;
- LocalAbstractObject medoid = null;
-
- for (int i = 0; i < idxs.length; i++) {
- float sum = shared.sumArray(idxs[i], idxs);
- if (sum < min) {
- min = sum;
- medoid = lst.get(i);
- }
- }
-
- return medoid;
- }
-
- /** Get furthest object in the data */
- protected LocalAbstractObject getFurthest() {
- float max = Float.MIN_VALUE;
- int maxIdx = 0;
- LocalAbstractObject f = null;
-
- while (true) {
- float[] dists = shared.getDistances(maxIdx);
- maxIdx = shared.maxDistInArray(dists);
- if (dists[maxIdx] > max) {
- f = shared.getObject(maxIdx);
- max = dists[maxIdx];
- } else {
- break; // we have found it
- }
- }
- return f;
- }
-
- /** Get furthest idx of object within given indexes
- * some object has to be the furthest.. else the method would not be called **/
- protected int getFurthestIdx(BitSet notUsedIdxs) {
- return getFurthestIdx(shared.distances, notUsedIdxs);
- }
-
- /** Get furthest idx of object within given indexes with respect to provided distMatrix
- * some object has to be the furthest.. else the method would not be called **/
- protected int getFurthestIdx(float[][] distMatrix, BitSet notUsedIdxs) {
- float max = Float.MIN_VALUE;
- int maxIdx = notUsedIdxs.nextSetBit(0); // first not used index
-
- while (true) {
- float[] dists = distMatrix[maxIdx];
- int candidateMaxIdx = shared.maxDistInArray(dists, notUsedIdxs); // max dist within the notUsedIdxs in array
- if (dists[candidateMaxIdx] > max) {
- max = dists[candidateMaxIdx];
- maxIdx = candidateMaxIdx;
- }
- else {
- break; // we have found it, thus return the idx from prev. iteration
- }
- }
- return maxIdx;
- }
-
-
- /**
- * Get minimum distance among the passed objects.
- * @param lst object to search in
- * @return minimum distance, zero if the passed list is empty or contains just one object.
- */
- protected float getMinDist(List<LocalAbstractObject> lst) {
- if (lst == null || lst.size() <= 1)
- return 0f;
- int[] idxs = shared.getIndexes(lst);
-
- float min = Float.MAX_VALUE;
- for (int i = 0; i < idxs.length - 1; i++) {
- float d = shared.minDist(idxs[i], idxs);
- if (d < min)
- min = d;
- }
- return min;
- }
-
- /** Find the maximum distance in the data ("shared" matrix) from the passed object.
- * @param objIdx index to {@link #shared} that specifies the object
- * @return max distance from the passed object to any other object in the data.
- */
- protected float getMaxDist(int objIdx) {
- float[] dists = shared.getDistances(objIdx);
- float max = dists[0];
- for (float d : dists) {
- if (d > max)
- max = d;
- }
- return max;
- }
-
- protected float getRadius(LocalAbstractObject center, List<LocalAbstractObject> lst) {
- float radius = 0f;
- int centerIdx = shared.getIndex(center);
- final float[] centerDists = shared.getDistances(centerIdx);
-
- if (lst == getObjects()) {
- for (float d : centerDists)
- if (d > radius)
- radius = d;
- } else {
- for (LocalAbstractObject o : lst) {
- float d = centerDists[shared.getIndex(o)];
- if (d > radius)
- radius = d;
- }
- }
-
- return radius;
- }
-
- /** Find all objects within the passed ball.
- * Its center is not included in the answer!
- * @param center ball center
- * @param radius ball radius
- * @return list of objects in the ball except the center object.
- */
- protected List<LocalAbstractObject> getBallObjects(LocalAbstractObject center, float radius, boolean excludeCovered) {
- List<LocalAbstractObject> res = new ArrayList<>();
- int centerIdx = shared.getIndex(center);
- float[] dists = shared.getDistances(centerIdx);
- for (int i = 0; i < shared.getObjectCount(); i++) {
- if (excludeCovered && isCovered(i))
- continue;
- if (dists[i] <= radius && centerIdx != i)
- res.add(shared.getObject(i));
- }
- return res;
- }
-
- /**
- * Find k-nearest neighbors to the passed object. This object itself is not returned as one of the neighbors.
- * @param center center (query object)
- * @param k number of nearest neighbors to indentify
- * @return list of k-nearest neighbor objects except the center object.
- */
- protected List<LocalAbstractObject> getKNNObjects(LocalAbstractObject center, int k) {
- int centerIdx = shared.getIndex(center);
- Ranked[] dists = Ranked.create(shared.getDistances(centerIdx));
- Arrays.sort(dists);
- List<LocalAbstractObject> res = new ArrayList<>(k);
- for (int i = 0; i < 1 + Math.min(k, dists.length); i++) {
- if (dists[i].index == centerIdx)
- continue;
- res.add(shared.getObject(dists[i].index));
- }
- return res;
- }
-
- /**
- * Find k-nearest neighbors to the passed object within the passed data objects. The center object itself is not returned as one of the neighbors.
- * All objects are assumed to be part of data in {@link #shared} data-set.
- * @param center center (query object)
- * @param k number of nearest neighbors to indentify
- * @param data list of objects to find neighbors in
- * @return list of k-nearest neighbor objects except the center object.
- */
- protected List<LocalAbstractObject> getKNNObjects(LocalAbstractObject center, int k, List<LocalAbstractObject> data) {
- int centerIdx = shared.getIndex(center);
- float[] dists = shared.getDistances(centerIdx);
-
- Ranked[] ranked = new Ranked[data.size()];
- for (int i = 0; i < data.size(); i++) {
- int objIdx = shared.getIndex(data.get(i));
- ranked[i] = new Ranked(dists[objIdx], i);
- }
- Arrays.sort(ranked);
- List<LocalAbstractObject> res = new ArrayList<>(k);
- for (int i = 0; i < Math.min(k, ranked.length); i++) {
- //if (ranked[i].index == centerIdx) // This is TODO!
- // continue;
- res.add(data.get(ranked[i].index));
- }
- return res;
- }
-
- /**
- * Find k-nearest neighbors to the passed object within the passed data objects.
- * The center object itself is not returned as one of the neighbors if it is not part of the passed dataIdxs array.
- * All objects are assumed to be part of data in {@link #shared} data-set.
- * @param centerIdx index of center (query object)
- * @param k number of nearest neighbors to indentify
- * @param dataIdxs array of indexes to objects where the neighbors are located in
- * @return list of indexes to k-nearest neighbor objects.
- */
- protected int[] getKNNIndexes(int centerIdx, int k, int[] dataIdxs) {
- float[] dists = shared.getDistances(centerIdx);
- return getKNNIndexesShared(k, dataIdxs, dists);
- }
-
- // centerPoint is outside the precomputed distances
- public int[] getKNNIndexesInOtherData(float[] dists, int k, int[] dataIdxs) {
- return getKNNIndexesShared(k, dataIdxs, dists);
- }
-
- protected int[] getKNNIndexes(int centerIdx, int k, BitSet idxs) {
- Ranked[] ranked = getRankedIdxs(idxs, shared.getDistances(centerIdx));
- int len = Math.min(k, ranked.length);
- int[] res = new int[len];
- for (int i = 0; i < len; i++)
- res[i] = ranked[i].index;
- return res;
-
- }
-
- private int[] getKNNIndexesShared(int k, int[] dataIdxs, float[] dists){
- Ranked[] ranked = getRankedIdxs(dataIdxs, dists); // sorted
- int len = Math.min(k, ranked.length);
- int[] res = new int[len];
- for (int i = 0; i < len; i++)
- res[i] = ranked[i].index;
- return res;
- }
-
- protected Ranked[] getRankedIdxs(int[] dataIdxs, float[] dists){
- Ranked[] ranked = new Ranked[dataIdxs.length];
- for (int i = 0; i < dataIdxs.length; i++) {
- int objIdx = dataIdxs[i];
- ranked[i] = new Ranked(dists[objIdx], i);
- }
- Arrays.sort(ranked);
- return ranked;
- }
-
- protected Ranked[] getRankedIdxs(BitSet dataIdxs, float[] dists){
- Ranked[] ranked = new Ranked[dataIdxs.cardinality()];
- int i = 0;
- for (int index = dataIdxs.nextSetBit(0); index >= 0;
- index = dataIdxs.nextSetBit(index + 1)) {
- ranked[i] = new Ranked(dists[index], index);
- i++;
- }
- Arrays.sort(ranked);
- return ranked;
- }
-
- public abstract String toShortString();
-
- public void write(OutputStream out) throws IOException {
- }
-
- /**
- * Selected representatives.
- * @return iterator over selected representatives.
- */
- public abstract Iterator<LocalAbstractObject> getRepresentatives();
-
- /**
- * Count of objects selected as representatives of the passed data
- * @return number of objects that are representatives
- */
- public abstract int getRepresentativesCount();
-
- public List<LocalAbstractObject> getObjects() {
- return shared.getObjects();
- }
-
- //****************************************************************************
- // HELPERS
- //****************************************************************************
-
- public static class Ranked implements Comparable<Ranked> {
- public final float distance;
- public final int index;
-
- public Ranked(float distance, int index) {
- this.distance = distance;
- this.index = index;
- }
-
- @Override
- public int compareTo(Ranked o) {
- int cmp = Float.compare(this.distance, o.distance);
- if (cmp == 0)
- cmp = Integer.compare(this.index, o.index);
- return cmp;
- }
-
- public static Ranked[] create(float[] dists) {
- Ranked[] ret = new Ranked[dists.length];
- for (int i = 0; i < dists.length; i++)
- ret[i] = new Ranked(dists[i], i);
- return ret;
- }
- }
-
- /** Class encapsulating data objects and precomputed distances related to them.
- * This is good for re-using the common work.
- */
- public static class PrecomputedDistances {
- /** Number of threads to use to compute the distance matrix -- default is one */
- public static int COMPUTATION_THREADS = 1;
-
- /** Data to summarize / represent */
- protected List<LocalAbstractObject> data;
-
- /** Map from object's locator to index in data or precomputed distances array */
- private Map<String,Integer> indexes;
- /** Distance matrix */
- private float[][] distances;
-
- public PrecomputedDistances(List<LocalAbstractObject> data) {
- this.data = data;
- // Precompute distance matrix
- initDistanceMatrix();
- }
-
- private void initDistanceMatrix() {
- long timeStart = System.currentTimeMillis();
-
- indexes = new HashMap<>(data.size());
- for (int i = 0; i < data.size(); i++)
- indexes.put(data.get(i).getLocatorURI(), i);
-
- if (COMPUTATION_THREADS == 1)
- initDistanceMatrixOneThread();
- else
- initDistanceMatrixMultiThread();
-
- long timeStop = System.currentTimeMillis();
- System.err.println("Distance matrix initialization for " + data.size() + " objects took: " + (timeStop - timeStart) + " msec");
- }
-
- private void initDistanceMatrixOneThread() {
- distances = new float[data.size()][data.size()];
- for (int i = 0; i < data.size(); i++) {
- //Arrays.fill(distances[i], 0);
- for (int j = i + 1; j < data.size(); j++) {
- float d = data.get(i).getDistance(data.get(j));
-// if (d == 0f)
-// throw new RuntimeException("Zero distance between " + data.get(i) + " and " + data.get(j));
- distances[i][j] = d;
- distances[j][i] = d;
- }
- }
- }
-
- private void initDistanceMatrixMultiThread() {
- boolean statsEnable = Statistics.isEnabledGlobally();
- Statistics.disableGlobally();
- distances = new float[data.size()][data.size()];
- try {
- long distsPerThread = data.size() / 2 * data.size() / COMPUTATION_THREADS;
- int poolThreads = (distsPerThread <= COMPUTATION_THREADS) ? 1 : COMPUTATION_THREADS;
- final ExecutorService pool = Executors.newFixedThreadPool(poolThreads);
-
- int[] counts = new int[data.size()];
- for (int i = 0; i < data.size(); i++) {
- counts[i] = data.size() - i + 1;
- }
-
- int totForThread = 0;
- int startForThread = 0;
- for (int i = 0; i < data.size(); i++) {
- totForThread += counts[i];
- if (totForThread >= distsPerThread) {
- pool.submit(new DistanceMatrixTask(data, distances, startForThread, i + 1));
- totForThread = 0;
- startForThread = i + 1;
- }
- }
- if (totForThread > 0)
- pool.submit(new DistanceMatrixTask(data, distances, startForThread, data.size()));
-
- pool.shutdown();
- while (!pool.awaitTermination(1, TimeUnit.MINUTES))
- ;
- } catch (InterruptedException ex) {
- Logger.getLogger(AbstractRepresentation.class.getName()).log(Level.SEVERE, null, ex);
- }
- if (statsEnable)
- Statistics.enableGlobally();
- }
-
- public int getObjectCount() {
- return data.size();
- }
-
- public LocalAbstractObject getObject(int i) {
- return data.get(i);
- }
-
- public List<LocalAbstractObject> getObjects() {
- return data;
- }
-
- public int getIndex(LocalAbstractObject o) {
- return indexes.get(o.getLocatorURI());
- }
-
- public int[] getIndexes(List<LocalAbstractObject> lst) {
- int[] idxs = new int[lst.size()];
- for (int i = 0; i < lst.size(); i++) {
- idxs[i] = indexes.get(lst.get(i).getLocatorURI());
- }
- return idxs;
- }
-
- public float getDistance(LocalAbstractObject o1, LocalAbstractObject o2) {
- return distances[indexes.get(o1.getLocatorURI())][indexes.get(o2.getLocatorURI())];
- }
-
- public float getDistance(int idx1, int idx2) {
- return distances[idx1][idx2];
- }
-
- public float[] getDistances(int objIdx) {
- return distances[objIdx];
- }
-
- public float[][] getDistances() {
- return distances;
- }
-
- public float sumArray(float[] arr) {
- float sum = 0;
- for (float v : arr)
- sum += v;
- return sum;
- }
-
- public float sumArray(float[] arr, int[] selectedIndexes) {
- float sum = 0;
- for (int i : selectedIndexes)
- sum += arr[i];
- return sum;
- }
-
- public float sumArray(int fromObjectAtIndex, int[] selectedIndexes) {
- return sumArray(distances[fromObjectAtIndex], selectedIndexes);
- }
-
- /** Index of maximum distance in the passed array */
- public int maxDistInArray(float[] arr) {
- int idx = -1;
- float max = 0f;
- for (int i = 0; i < arr.length; i++) {
- if (idx == -1 || max < arr[i]) {
- max = arr[i];
- idx = i;
- }
- }
- return idx;
- }
-
- /** Index of maximum distance in the passed array considering only not used indexes */
- public int maxDistInArray(float[] arr, BitSet notUsedIndexes) {
- int idx = -1;
- float max = 0f;
- for (int i = notUsedIndexes.nextSetBit(0); i >= 0;
- i = notUsedIndexes.nextSetBit(i + 1)) {
- if (idx == -1 || max < arr[i]) {
- max = arr[i];
- idx = i;
- }
- }
- return idx;
- }
-
- public int minDistInArray(float[] arr, BitSet notUsedIndexes) {
- int idx = -1;
- float min = Float.MAX_VALUE;
- for (int i = notUsedIndexes.nextSetBit(0); i >= 0;
- i = notUsedIndexes.nextSetBit(i + 1)) {
- if (idx == -1 || min > arr[i]) {
- min = arr[i];
- idx = i;
- }
- }
- return idx;
- }
-
- public int minDistInArrayExceptIdx(float[] arr, BitSet notUsedIndexes, int notIncludeIdx) {
- int idx = -1;
- float min = Float.MAX_VALUE;
- for (int i = notUsedIndexes.nextSetBit(0); i >= 0;
- i = notUsedIndexes.nextSetBit(i + 1)) {
- if(i == notIncludeIdx){
- continue;
- }
- if (idx == -1 || min > arr[i]) {
- min = arr[i];
- idx = i;
- }
- }
- return idx;
- }
-
- public float minDist(int fromObjectAtIndex, int[] selectedIndexes) {
- return minDist(distances[fromObjectAtIndex], fromObjectAtIndex, selectedIndexes);
- }
-
- public MinDistObjectPair minDist(BitSet selectedIndexes) {
- float min = Float.MAX_VALUE;
- int minIndexFst = -1;
- int minIndexSec = -1;
- for (int i = selectedIndexes.nextSetBit(0); i >= 0;
- i = selectedIndexes.nextSetBit(i + 1)) {
- for (int j = selectedIndexes.nextSetBit(i + 1); j >= 0;
- j = selectedIndexes.nextSetBit(j + 1)) {
- float d = distances[i][j];
- if (d < min) {
- min = d;
- minIndexFst = i;
- minIndexSec = j;
- }
- }
- }
- return new MinDistObjectPair(minIndexFst, minIndexSec, min);
- }
-
- public MinDistObjectPair minDist(int[] selectedIndexes) {
- return minDist(distances, selectedIndexes);
- }
-
- public int minDistIdx(float[] arr, int fromObjectAtIndex){
- float min = Float.MAX_VALUE;
- int idx = fromObjectAtIndex;
- for (int i = 0; i < arr.length; i++) {
- if (i == fromObjectAtIndex)
- continue; // skip myself
- float d = arr[i];
- if (d < min) {
- min = d;
- idx = i;
- }
- }
- return idx;
- }
-
- public float minDist(float[] arr, int fromObjectAtIndex, int[] selectedIndexes) {
- float min = Float.MAX_VALUE;
- for (int i = 0; i < selectedIndexes.length; i++) {
- if (selectedIndexes[i] == fromObjectAtIndex)
- continue; // skip myself
- float d = arr[selectedIndexes[i]];
- if (d < min)
- min = d;
- }
- return min;
- }
-
-
- /** Thread for computing a new artifical cluster center as a mean vector of all passed vectors. */
- private class DistanceMatrixTask implements Runnable {
- private final List<LocalAbstractObject> data;
- private final float[][] distances;
- private final int start;
- private final int end;
-
- public DistanceMatrixTask(List<LocalAbstractObject> data, float[][] distances, int indexStart, int indexEnd) {
- this.data = data;
- this.distances = distances;
- this.start = indexStart;
- this.end = indexEnd;
- }
-
- @Override
- public void run() {
- for (int i = start; i < end; i++) {
- for (int j = i + 1; j < data.size(); j++) {
- float d = data.get(i).getDistance(data.get(j));
- distances[i][j] = d;
- distances[j][i] = d;
- }
- }
- }
- }
-
- public MinDistObjectPair minDist(float[][] arr, int[] selectedIndexes) {
- float min = Float.MAX_VALUE;
- int minIndexFst = -1;
- int minIndexSec = -1;
- for (int i = 0; i < selectedIndexes.length; i++) {
- for (int j = i + 1; j < selectedIndexes.length; j++) {
- float d = arr[i][j];
- if (d < min) {
- min = d;
- minIndexFst = selectedIndexes[i];
- minIndexSec = selectedIndexes[j];
- }
- }
- }
- return new MinDistObjectPair(minIndexFst, minIndexSec, min);
- }
-
- /** Computes distances from an extObject to all data objects */
- public float[] getDistancesToExtObj(LocalAbstractObject extObject){
- float[] dists = new float[data.size()];
- for (int i = 0; i < data.size(); i++)
- dists[i] = extObject.getDistance(data.get(i));
- return dists;
- }
-
- }
-
- public static class MinDistObjectPair implements Comparable<MinDistObjectPair>{
- protected int firstIdx;
- protected int secondIdx;
- protected float distance;
-
- public MinDistObjectPair(int indexFst, int indexSec, float distance){
- this.distance = distance;
- this.firstIdx = indexFst;
- this.secondIdx = indexSec;
- }
-
- @Override
- public int compareTo(MinDistObjectPair o) {
- int cmp = Float.compare(this.distance, o.distance);
- if (cmp == 0)
- cmp = Integer.compare(this.firstIdx, o.firstIdx);
- if (cmp == 0)
- cmp = Integer.compare(this.secondIdx, this.secondIdx);
- return cmp;
- }
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/BallRepresentation.java b/src/cz/muni/fi/disa/similarityoperators/cover/BallRepresentation.java
deleted file mode 100644
index 40a7321..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/BallRepresentation.java
+++ /dev/null
@@ -1,110 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import cz.muni.fi.disa.similarityoperators.cover.types.BallBasedCover;
-import messif.objects.LocalAbstractObject;
-import messif.objects.keys.RadiusObjectKey;
-import messif.utility.FileUtils;
-
-import java.io.IOException;
-import java.util.Collections;
-import java.util.Iterator;
-import java.util.List;
-
-/**
- * Ball representation of data set.
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class BallRepresentation extends AbstractRepresentation<BallRepresentation> implements BallBasedCover {
- private LocalAbstractObject center;
- private float radius;
-
- /** Instantiate a ball representation from a serialized file containing one object the key of which is RadiusObjectKey that holds the ball radius.
- * @param locatorID ID/symbolic name of this representation
- * @param objCls class of objects
- * @param filename file containing MESSIF serialized objects
- * @throws IllegalArgumentException
- * @throws IOException
- */
- public BallRepresentation(String locatorID, Class<LocalAbstractObject> objCls, String filename) throws IllegalArgumentException, IOException {
- super(locatorID, Collections.EMPTY_LIST);
- center = FileUtils.readObjects(objCls, filename).get(0);
- radius = center.getObjectKey(RadiusObjectKey.class).getRadius();
- }
-
- public BallRepresentation(List<LocalAbstractObject> data) {
- super(data);
- }
-
- public BallRepresentation(PrecomputedDistances shared) {
- super(shared);
- }
-
- public void setCenter(LocalAbstractObject center) {
- this.center = center;
- radius = getRadius(center, shared.getObjects());
- setCoveredByBall(center, radius);
- }
-
- /** Caution: this method just sets the radius manually and recalculated coverage of data in shared array. */
- public void setRadius(float radius) {
- this.radius = radius;
- setCoveredByBall(center, radius);
- }
-
- @Override
- protected void select() {
- setCenter(getMedoid());
- }
-
- @Override
- public boolean isExternalCovered(LocalAbstractObject o) {
- return (o.getDistance(center) <= radius);
- }
-
- public CoverRank getExternalCoverRank(LocalAbstractObject o) {
- final float d = o.getDistance(center);
- return new CoverRank((d <= radius), d);
- }
-
- @Override
- public String toString() {
- StringBuilder sb = new StringBuilder(getClass().getSimpleName());
- sb.append('[');
- sb.append("objects=").append(shared.getObjectCount()).append(", ");
- sb.append("center=").append(center).append(", radius=").append(radius).append(']');
- return sb.toString();
- }
-
- @Override
- public String toShortString() {
- StringBuilder sb = new StringBuilder(getClass().getSimpleName());
- sb.append('[');
- if (locatorID != null)
- sb.append("id=").append(locatorID).append(", ");
- sb.append("objects=").append(shared.getObjectCount()).append(", ");
- sb.append("radius=").append(radius).append(']');
- return sb.toString();
- }
-
- @Override
- public Iterator<LocalAbstractObject> getRepresentatives() {
- return Collections.singletonList(center).iterator();
- }
-
- @Override
- public int getRepresentativesCount() {
- return 1;
- }
-
- @Override
- public float getAverageRadius() {
- return radius;
- }
-
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/CoverRank.java b/src/cz/muni/fi/disa/similarityoperators/cover/CoverRank.java
deleted file mode 100644
index 4f6c74c..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/CoverRank.java
+++ /dev/null
@@ -1,42 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.cover;
-
-/**
- * Encapsulated result of coverage test and ranking of the object, i.e. how much the object is contained in a representation (aka in the middle of a cluster).
- * The smaller value of rank, the better.
- */
-public class CoverRank implements Comparable<CoverRank> {
- boolean covered;
- float rank;
-
- public CoverRank(boolean covered, float rank) {
- this.covered = covered;
- this.rank = rank;
- }
-
- @Override
- public int compareTo(CoverRank o) {
- if (this.covered != o.covered) {
- return (o.covered) ? 1 : -1;
- }
- return Float.compare(this.rank, o.rank); // The lower the rank the better (more in the middle of the hull)
- }
-
- public boolean isCovered() {
- return covered;
- }
-
- public float getRank() {
- return rank;
- }
-
- @Override
- public String toString() {
- return "CoverRank{" + "covered=" + covered + ", rank=" + rank + '}';
- }
-
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/FVDMRepresentation.java b/src/cz/muni/fi/disa/similarityoperators/cover/FVDMRepresentation.java
deleted file mode 100644
index 7278a47..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/FVDMRepresentation.java
+++ /dev/null
@@ -1,84 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.cover;
-
-
-import messif.objects.LocalAbstractObject;
-import messif.utility.FileUtils;
-
-import java.io.IOException;
-import java.util.List;
-
-/**
- * FVDM method for hull vertices (points) -- full name: Finding Vertices with Distance Matrix
- *
- * This variant does not use any out-of-sample data to select the hull point finely.
- *
- * Paper: Zhong, J., Tang, K., & Qin, A. K. (2014). Finding convex hull vertices in metric space.
- * In Proceedings of the International Joint Conference on Neural Networks (pp. 1587–1592).
- * Institute of Electrical and Electronics Engineers Inc. https://doi.org/10.1109/IJCNN.2014.6889699
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class FVDMRepresentation extends AbstractHullRepresentation<FVDMRepresentation> {
-
- /** Instantiate a FVDM representation from a serialized file containing the outlier objects.
- * @param locatorID ID/symbolic name of this representation
- * @param objCls class of objects
- * @param filename file containing MESSIF serialized objects
- * @throws IllegalArgumentException
- * @throws IOException
- */
- public FVDMRepresentation(String locatorID, Class<LocalAbstractObject> objCls, String filename) throws IllegalArgumentException, IOException {
- super(locatorID, FileUtils.readObjects(objCls, filename)); // We do not select hull objects here, we rather take existing ones (already selected).
- hull.addAll(getObjects());
- }
-
- public FVDMRepresentation(List<LocalAbstractObject> data) {
- super(data);
- }
-
- public FVDMRepresentation(PrecomputedDistances shared) {
- super(shared);
- }
-
- @Override
- protected void select() {
- fvdmSelect();
- //fvdmSelectByPaper();
- setCoveredByHull();
- }
-
- private void fvdmSelect() {
- for (int i = 0; i < shared.getObjectCount(); i++) {
- float[] dists = shared.getDistances(i);
- float max = dists[0];
- int maxIdx = 0;
- for (int j = 1; j < shared.getObjectCount(); j++) {
- float d = dists[j];
- if (d > max) {
- max = d;
- maxIdx = j;
- }
- }
- LocalAbstractObject hoNew = shared.getObject(maxIdx);
- if (!hull.contains(hoNew))
- hull.add(hoNew);
- }
- }
-
- private void fvdmSelectByPaper() {
- for (int i = 0; i < shared.getObjectCount(); i++) {
- for (int j = 0; j < shared.getObjectCount(); j++) {
- float maxFromJ = getMaxDist(j);
- if (maxFromJ == shared.getDistance(i, j)) {
- hull.add(shared.getObject(i));
- break;
- }
- }
- }
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/HullBorderRepresentation.java b/src/cz/muni/fi/disa/similarityoperators/cover/HullBorderRepresentation.java
deleted file mode 100644
index a101151..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/HullBorderRepresentation.java
+++ /dev/null
@@ -1,142 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import messif.objects.LocalAbstractObject;
-
-import java.util.Arrays;
-import java.util.List;
-import java.util.Map;
-import java.util.TreeMap;
-
-/**
- * BORDER method for hull vertices (points) -- BOundaRy points DEtectoR
- *
- * It employs the special property of the reverse k nearest neighbor (RkNN).
- * The boundary points whose RkNN number is smaller than a user predefined threshold can be output incrementally.
- *
- * Here, we select boundary objects while coverage by the hull objects is increasing and we also use the optimization procedure from our Hull
- * to minimize the number of hull objects.
- *
- * Paper: Xia, C., Hsu, W., Lee, M. L., & Ooi, B. C. (2006).
- * BORDER: Efficient computation of boundary points.
- * IEEE Transactions on Knowledge and Data Engineering, 18(3), 289–303. https://doi.org/10.1109/TKDE.2006.38
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class HullBorderRepresentation extends AbstractHullRepresentation<HullFastBorderRepresentation> {
- /** Ration of kThreshold / K used by the authors. In particular, it was 35 / 50. */
- private static final float K_TO_THRESHOLD_RATIO = 0.70f;
-
- /** First values is object index; second value is rnnCount */
- private int[][] rnnCounts;
-
- public HullBorderRepresentation(List<LocalAbstractObject> data) {
- super(data);
- }
-
- public HullBorderRepresentation(PrecomputedDistances shared) {
- super(shared);
- }
-
- @Override
- protected void select() {
- // Take 1/4 of data at maximum!
- int k = (shared.getObjectCount() >= 200) ? 50 : shared.getObjectCount() / 4;
- int kThreshold = (int)(K_TO_THRESHOLD_RATIO * k); // was: 35;
-
- initReverseNNCounts(k, kThreshold);
- // Take the first 3
- hull.add(shared.getObject(rnnCounts[0][0]));
- hull.add(shared.getObject(rnnCounts[1][0]));
- hull.add(shared.getObject(rnnCounts[2][0]));
-
- setCoveredByHull();
- float bestCoverage = getCoverage();
-
- int prevRnnCount = 0;
- for (int i = 3; !isAllCovered() && i < rnnCounts.length && rnnCounts[i][1] < kThreshold; i++) {
- if (rnnCounts[i][1] != prevRnnCount) {
- prevRnnCount = rnnCounts[i][1];
- System.err.println("Current RkNN value: " + prevRnnCount + " Threshold: " + kThreshold + " Obj index: " + i);
- }
- LocalAbstractObject o = shared.getObject(i);
- hull.add(o);
- setCoveredByHull();
- float coverage = getCoverage();
- if (coverage >= bestCoverage) {
- bestCoverage = coverage;
- System.err.println("Current coverage: " + bestCoverage + " Hull size: " + hull.size());
- } else {
- bestCoverage = optimize(bestCoverage);
- }
- }
- setCoveredByHull();
- System.err.println("Border hull: " + hull.size() + " objects");
- System.err.println("Coverage: " + (int)Math.round(getCoverage() * 100f) + "%");
- System.err.println("Cov ratio: " + getCoverageScore());
- }
-
- /**
- * Taken from {@link HullOptimizedRepresentationV2} but the "presenceInHull" is removed.
- * @return resulting coverage after optimization
- */
- private float optimize(float coverageBest) {
- int indexBest = -1;
-
- for (int i = 0; i < hull.size() - 1; i++) {
- LocalAbstractObject hoRemoved = hull.remove(i);
- setCoveredByHull();
- float cvg = getCoverage();
- if (cvg > coverageBest) {
- coverageBest = cvg;
- indexBest = i;
- }
- hull.add(i, hoRemoved);
- }
- // Improved by replacement, so apply the removal
- if (indexBest >= 0) {
- hull.remove(indexBest);
- // Update the coverage back to original state!
- setCoveredByHull();
- System.err.print("# Removed hull object at " + indexBest + " to increase coverage: "); System.err.println(toString());
- } else {
- setCoveredByHull(); // Restore original state
- }
- return coverageBest;
- }
-
- /** Calculate counts of being kNN for the passed "k" */
- private void initReverseNNCounts(int k, int kThreshold) {
- long start = System.nanoTime();
- rnnCounts = new int[shared.getObjectCount()][2];
- int[] allData = new int[shared.getObjectCount()];
- for (int i = 0; i < shared.getObjectCount(); i++) {
- allData[i] = i;
- rnnCounts[i] = new int[] {i, 0};
- }
- for (int i = 0; i < shared.getObjectCount(); i++) {
- int[] knns = getKNNIndexes(i, k, allData);
- for (int idx : knns)
- rnnCounts[idx][1]++;
- }
- Arrays.sort(rnnCounts, (int[] o1, int[] o2) -> {
- return Integer.compare(o1[1], o2[1]);
- });
- System.err.println("RnnCounts took: " + (System.nanoTime() - start) / 10000000f + " msec");
- // Get histogram of RkNN counts
- Map<Integer,Integer> hist = new TreeMap<>();
- int objsUpToThreshold = 0;
- for (int[] cnt : rnnCounts) {
- hist.put(cnt[1], hist.getOrDefault(cnt[1], 0) + 1);
- if (cnt[1] < kThreshold)
- objsUpToThreshold++;
- }
- System.err.println("RkNN counts histogram: " + hist);
- System.err.println("Objects up to kThreshold " + kThreshold + ": " + objsUpToThreshold);
- }
-
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/HullCenterRepresentation.java b/src/cz/muni/fi/disa/similarityoperators/cover/HullCenterRepresentation.java
deleted file mode 100644
index efc4b63..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/HullCenterRepresentation.java
+++ /dev/null
@@ -1,45 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import messif.objects.LocalAbstractObject;
-
-import java.util.List;
-
-/**
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class HullCenterRepresentation extends HullRepresentation {
- /** Center of the hull -- can be selected from covered objects or given from outside */
- private LocalAbstractObject center;
-
- public HullCenterRepresentation(List<LocalAbstractObject> data) {
- super(data);
- }
-
- public HullCenterRepresentation(PrecomputedDistances shared) {
- super(shared);
- }
-
- public LocalAbstractObject getCenter() {
- return center;
- }
-
- public void setCenter(LocalAbstractObject center) {
- this.center = center;
- }
-
- @Override
- public CoverRank getExternalCoverRank(LocalAbstractObject o) {
- CoverRank rank = super.getExternalCoverRank(o);
- float d = o.getDistance(center);
- System.err.println("HullRank: ID=" + getLocatorID() + ", hull_center=" + center.getLocatorURI() + ", obj=" + o.getLocatorURI()
- + ", hullrank=" + rank.rank + ", centerrank=" + d + ",covered=" + ((rank.isCovered()) ? 1 : 0 ));
- rank.rank = d; // Just update the rank here...
- return rank;
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/HullFastBorderRepresentation.java b/src/cz/muni/fi/disa/similarityoperators/cover/HullFastBorderRepresentation.java
deleted file mode 100644
index 7b4b3d9..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/HullFastBorderRepresentation.java
+++ /dev/null
@@ -1,111 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import messif.objects.LocalAbstractObject;
-
-import java.util.*;
-
-/**
- * BORDER method for hull vertices (points) -- BOundaRy points DEtectoR
- *
- * It employs the special property of the reverse k nearest neighbor (RkNN).
- * The boundary points whose RkNN number is smaller than a user predefined threshold can be output incrementally.
- *
- * Here, we select boundary objects in the order of increasing RkNN count but only up to a predefined number of objects.
- * The best hull in terms of coverage is output.
- *
- * Paper: Xia, C., Hsu, W., Lee, M. L., & Ooi, B. C. (2006).
- * BORDER: Efficient computation of boundary points.
- * IEEE Transactions on Knowledge and Data Engineering, 18(3), 289–303. https://doi.org/10.1109/TKDE.2006.38
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class HullFastBorderRepresentation extends AbstractHullRepresentation<HullFastBorderRepresentation> {
- /** Ration of kThreshold / K used by the authors. In particular, it was 35 / 50. */
- private static final float K_TO_THRESHOLD_RATIO = 0.70f;
-
- public static final int MAX_HULL_SIZE = 150;
-
- /** First values is object index; second value is rnnCount */
- private int[][] rnnCounts;
-
- public HullFastBorderRepresentation(List<LocalAbstractObject> data) {
- super(data);
- }
-
- public HullFastBorderRepresentation(PrecomputedDistances shared) {
- super(shared);
- }
-
- @Override
- protected void select() {
- // Take 1/4 of data at maximum!
- int k = (shared.getObjectCount() >= 200) ? 50 : shared.getObjectCount() / 4;
- int kThreshold = (int)(K_TO_THRESHOLD_RATIO * k); // was: 35;
-
- initReverseNNCounts(k, kThreshold);
- // Take the first 3
- hull.add(shared.getObject(rnnCounts[0][0]));
- hull.add(shared.getObject(rnnCounts[1][0]));
- hull.add(shared.getObject(rnnCounts[2][0]));
-
- setCoveredByHull();
- float bestScore = getCoverageScore();
- int bestIndex = 2;
-
- for (int i = 3; !isAllCovered() && i < rnnCounts.length && rnnCounts[i][1] < kThreshold && i < MAX_HULL_SIZE; i++) {
- LocalAbstractObject o = shared.getObject(i);
- hull.add(o);
- setCoveredByHull();
- float score = getCoverageScore();
- if (score >= bestScore) {
- bestScore = score;
- bestIndex = i;
- }
- }
- // Remove all hull objects until we get the best configuration.
- while ((hull.size() - 1) > bestIndex)
- hull.remove(hull.size() - 1);
-
- setCoveredByHull();
- System.err.println("Border hull: " + hull.size() + " objects");
- System.err.println("Coverage: " + (int)Math.round(getCoverage() * 100f) + "%");
- System.err.println("Cov ratio: " + getCoverageScore());
- }
-
- /** Calculate counts of being kNN for the passed "k" */
- private void initReverseNNCounts(int k, int kThreshold) {
- long start = System.nanoTime();
- rnnCounts = new int[shared.getObjectCount()][2];
- int[] allData = new int[shared.getObjectCount()];
- for (int i = 0; i < shared.getObjectCount(); i++) {
- allData[i] = i;
- rnnCounts[i] = new int[] {i, 0};
- }
- for (int i = 0; i < shared.getObjectCount(); i++) {
- int[] knns = getKNNIndexes(i, k, allData);
- for (int idx : knns)
- rnnCounts[idx][1]++;
- }
- Collections.shuffle(Arrays.asList(hull)); // Try to have different ordering of objects of the same RkNN counts.
- Arrays.sort(rnnCounts, (int[] o1, int[] o2) -> {
- return Integer.compare(o1[1], o2[1]);
- });
- System.err.println("RnnCounts took: " + (System.nanoTime() - start) / 10000000f + " msec");
- // Get histogram of RkNN counts
- Map<Integer,Integer> hist = new TreeMap<>();
- int objsUpToThreshold = 0;
- for (int[] cnt : rnnCounts) {
- hist.put(cnt[1], hist.getOrDefault(cnt[1], 0) + 1);
- if (cnt[1] < kThreshold)
- objsUpToThreshold++;
- }
- System.err.println("RkNN counts histogram: " + hist);
- System.err.println("Objects up to kThreshold " + kThreshold + ": " + objsUpToThreshold);
- }
-
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/HullIncrementalRepresentation.java b/src/cz/muni/fi/disa/similarityoperators/cover/HullIncrementalRepresentation.java
deleted file mode 100644
index df219f0..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/HullIncrementalRepresentation.java
+++ /dev/null
@@ -1,429 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import cz.muni.fi.disa.similarityoperators.utils.MergingType;
-import messif.objects.LocalAbstractObject;
-
-import java.io.FileOutputStream;
-import java.io.IOException;
-import java.io.OutputStream;
-import java.util.*;
-
-/**
- * Incrementally build a hull over the passed dataset.
- *
- * Initial 3-object hulls are created in the order of closest objects (first,second) to which the third object (as 1NN(first) or 1NN(second)) is added.
- * Iteratively the initial hulls are merged -- the closest hulls first. The closest hulls HS to a hull H is obtained on hull objects, i.e.
- * the closest objects win (i.e. 1NN join(H,HS) is executed).
- * The iterations are "generation"-based -- all hulls are merged first, then such hulls are taken to the next iteration. The reduction in the number of hulls is 2.
- *
- * @author Miriama Janosova, Masaryk University, Brno, Czech Republic, x424615@fi.muni.cz
- */
-public class HullIncrementalRepresentation extends HullRepresentation {
-
- private List<HullRepresentation> hullArray;
- /** Pair-wise distances among hull objects */
- protected float[][] hullMinDistances;
- protected MergingType mergingType;
-
- public HullIncrementalRepresentation(List<LocalAbstractObject> data, MergingType type) {
- super(data);
- this.hullArray = new ArrayList<>();
- this.mergingType = type;
- }
-
- public HullIncrementalRepresentation(PrecomputedDistances shared, MergingType type) {
- super(shared);
- this.hullArray = new ArrayList<>();
- this.mergingType = type;
- }
-
- public HullIncrementalRepresentation(PrecomputedDistances shared) {
- super(shared);
- this.hullArray = new ArrayList<>();
- this.mergingType = MergingType.NONE;
- }
-
- public HullIncrementalRepresentation(List<LocalAbstractObject> data) {
- super(data);
- this.hullArray = new ArrayList<>();
- this.mergingType = MergingType.NONE;
- }
-
- @Override
- protected boolean initHullPoints() {
- if (shared.data.size() <= 3) {
- hull.addAll(shared.data);
- return false;
- }
- BitSet notProcessedIndexes = new BitSet(this.shared.getObjectCount());
- notProcessedIndexes.set(0, this.shared.getObjectCount()); //sets bits to 1
- MinDistObjectPairIterator iterator = new MinDistObjectPairIterator(shared.getDistances(), notProcessedIndexes); // updates notProcessedIndexes internally
-
- while (iterator.hasNext()) {
- if (notProcessedIndexes.cardinality() < 3) {
- for (int i = notProcessedIndexes.nextSetBit(0); i != -1;
- i = notProcessedIndexes.nextSetBit(i + 1)) { // find spare objects and merge them with closest hull
- addObjectToClosestHull(i); // i is index
- }
- return true; // finish
- }
- MinDistObjectPair minDistsObject = iterator.next();
- int third = findClosestThirdIdx(notProcessedIndexes, minDistsObject.firstIdx, minDistsObject.secondIdx);
- notProcessedIndexes.clear(third);
-
- List<LocalAbstractObject> hullPoints = new ArrayList<>(Arrays.asList(shared.getObject(minDistsObject.firstIdx), // vytvorim hull nad najdenymi bodmi
- shared.getObject(minDistsObject.secondIdx), shared.getObject(third)));
- HullRepresentation newHull = new HullRepresentation(hullPoints); //create new hull
- newHull.hull.addAll(hullPoints); // set hull points
- hullArray.add(newHull);
- }
- return true;
- }
-
- /** Identify an object that is the closest to the passed objects (all by indexes) for creating initial hulls.
- *
- * It pick the object that is either 1NN of the first passed object or 1NN of the second passed objects
- * depending on which 1NN is closer.
- *
- * @param notProcessed list of indexes of unused objects
- * @param firstIdx index of the first object
- * @param secondIdx index of the second object
- * @return index of an object closest to the passed ones.
- */
- protected int findClosestThirdIdx(BitSet notProcessed, int firstIdx, int secondIdx){
- int[] firstCandidate = getKNNIndexes(firstIdx, 1, notProcessed);
- int[] secondCandidate = getKNNIndexes(secondIdx, 1, notProcessed);
- int firstCandIdx = firstCandidate[0];
- int secondCandIdx = secondCandidate[0];
- if (this.shared.getDistance(firstIdx, firstCandIdx) <=
- this.shared.getDistance(secondIdx, secondCandIdx)){
- return firstCandIdx;
- }
- return secondCandIdx;
- }
-
- /** Find the closest hull to the passed object (by its index) and add the object to the hull.
- *
- * @param index index of object to find the closest hull for
- */
- protected void addObjectToClosestHull(int index) {
- int[] closesIdx = getKNNIndexes(index, 1, getAllCurrentHullIndexes()); // closest hull object among all hull objects of all hulls.
- LocalAbstractObject object = shared.getObject(index);
-
- HullRepresentation closestHull = hullArray.get(findCorrespondingHullIdx(shared.getObject(closesIdx[0])));
-
- closestHull.shared.data.add(object); // add just to data
- if (isCoveredByHull(index, closestHull.getRepresentativesCount(), shared.getIndexes(closestHull.hull))) {
- return; // don't enlarge the hull
- }
- closestHull.hull.add(object); //otherwise add to hull points
- }
-
- /** Get index of a hull object in the hulls */
- protected int findCorrespondingHullIdx(LocalAbstractObject object) {
- return findCorrespondingHullIdx(object, hullArray);
- }
-
- /** Get index of a hull object in the passed hulls */
- private int findCorrespondingHullIdx(LocalAbstractObject object, List<HullRepresentation> hulls){
- for (int i = 0; i < hulls.size(); i++) {
- if (hulls.get(i).hull.contains(object))
- return i;
- }
- return -1; // a hull has to be closest, does not make sense otherwise
- }
-
- protected List<LocalAbstractObject> getAllCurrentHullObjects() {
- return getAllHullObjects(this.hullArray);
- }
-
- private List<LocalAbstractObject> getAllHullObjects(List<HullRepresentation> hulls) {
- List<LocalAbstractObject> hullList = new ArrayList<>();
- for (HullRepresentation h : hulls) {
- hullList.addAll(h.hull);
- }
- return hullList;
- }
-
- protected int[] getAllCurrentHullIndexes() {
- return shared.getIndexes(getAllCurrentHullObjects());
- }
-
- private float[][] precomputeHullDistances(List<HullRepresentation> hullArray){
- float[][] minDistances = new float[hullArray.size()][hullArray.size()];
- for (int i = 0; i < hullArray.size(); i++) { // pre vsetky hulls
- for (int j = i + 1; j < hullArray.size(); j++) { // prejdi zvysne
- float minDist = Float.MAX_VALUE;// udrzi min vzdialenost medzi vsetkymi komb hullPointov
- for (LocalAbstractObject firstHP : hullArray.get(i).hull){ // prejdi vsetky hull points prveho a najdi najmensiu vzial k hociakemu druhemu
- for (LocalAbstractObject secondHP : hullArray.get(j).hull){
- if (shared.getDistance(firstHP, secondHP) < minDist){
- minDist = shared.getDistance(firstHP, secondHP);
- }
- }
- }
- if (minDist == 0f)
- throw new RuntimeException("Zero distance between " + hullArray.get(i).toString() +
- " and " + hullArray.get(j).toString());
- minDistances[i][j] = minDist;
- minDistances[j][i] = minDist;
- }
- }
- return minDistances;
- }
-
- protected void precomputeHullDistances(){
- hullMinDistances = precomputeHullDistances(hullArray);
- }
-
- @Override
- public void select() {
- if (!initHullPoints())
- return;
- System.out.println("First generation \n");
- for (int i = 0; i < hullArray.size(); i++) {
- try {
- OutputStream out = new FileOutputStream("./hullIncTest/pripad 3.1/generations/gen1/h" + i + ".txt", false);
- hullArray.get(i).write(out);
- } catch (IOException e) {
- e.printStackTrace();
- }
- }
-
- int generationNumber = 2;
- while (hullArray.size() != 1) {
-// System.out.println("NEW GENERATION of hulls" + generationNumber + "\n");
-// String path = "./hullIncTest/pripad 3.1/generations/gen" + generationNumber + "/";
- precomputeHullDistances();
- BitSet notProcessedIndexes = new BitSet(hullMinDistances.length);
- notProcessedIndexes.set(0, hullMinDistances.length); //sets bits to 1
- List<HullRepresentation> newHullArray = new ArrayList<>();
- MinDistObjectPairIterator iterator = new MinDistObjectPairIterator(hullMinDistances, notProcessedIndexes);
- int hullNum = 0;
-
- while (iterator.hasNext()) {
- if (notProcessedIndexes.cardinality() == 1) { // there is either one spare hull or none (even or odd number of them)
- int index = notProcessedIndexes.nextSetBit(0);
- int closestIdx = getClosetsHullIdx(index, newHullArray);// returns closest hull within newHullArray
- mergeHullsByIndexes(index, closestIdx, newHullArray);
- newHullArray.remove(closestIdx); // uprav newHullArray, aby obsahovalo vyhradne newHull
- break;
- }
- MinDistObjectPair minDistsObject = iterator.next();
- mergeHullsByIndexes(minDistsObject.firstIdx, minDistsObject.secondIdx, newHullArray);
- hullNum ++;
- }
- hullArray = newHullArray;
- generationNumber ++;
- }
- this.hull = hullArray.get(0).hull;
- setCoveredByHull();
- }
-
- /** get closest hull within hulls */
- private int getClosetsHullIdx(int hullIndex, List<HullRepresentation> hulls){
- List<LocalAbstractObject> allHulls = getAllHullObjects(hulls);
- LocalAbstractObject closestPoint = null;
- float minDist = Float.MAX_VALUE;
- HullRepresentation bbb = hullArray.get(hullIndex);
- List<LocalAbstractObject> uf = bbb.hull;
- for (LocalAbstractObject hullPoint : uf ){
- List<LocalAbstractObject> knn = getKNNObjects(hullPoint, 1, allHulls);
- if (this.shared.getDistance(knn.get(0), hullPoint) < minDist) {
- closestPoint = knn.get(0);
- minDist = shared.getDistance(closestPoint, hullPoint);
- }
- }
- return findCorrespondingHullIdx(closestPoint, hulls);
- }
-
- private List<LocalAbstractObject> getPointsToComputeCoverage(List<HullRepresentation> hulls,
- LocalAbstractObject excludedPoint,
- HashSet<LocalAbstractObject> covered) {
- List<LocalAbstractObject> hullPoints = new ArrayList<>();
- hulls.forEach(hull -> hullPoints.addAll(hull.hull));
-
- hullPoints.remove(excludedPoint);
- hullPoints.removeAll(covered);
-
- return hullPoints;
- }
-
- private List<LocalAbstractObject> getPointsToComputeCoverage(HullRepresentation first,
- HullRepresentation second,
- LocalAbstractObject notToIncludePoint,
- HashSet<LocalAbstractObject> covered){
- List<LocalAbstractObject> hullPoints = new ArrayList<>(first.hull);
- hullPoints.addAll(second.hull);
- hullPoints.remove(notToIncludePoint);
- hullPoints.removeAll(covered);
- return hullPoints;
- }
-
- @Override
- public boolean isCoveredByHull(int objIdx, int knHullPoints, int[] hullIdxs) {
- int[] knns = getKNNIndexes(objIdx, knHullPoints, hullIdxs);
- return isCoveredByDistSum(objIdx, hullIdxs[knns[0]], hullIdxs);
- }
-
- private void mergeHullsByIndexes(int firstIdx, int secondIdx, List<HullRepresentation> newHullArray){
- newHullArray.add(mergeHulls(hullArray.get(firstIdx), hullArray.get(secondIdx)));
- }
-
- protected HullRepresentation mergeHulls(List<HullRepresentation> hulls) {
- if (hulls.size() == 0) return null;
- if (hulls.size() == 1) return hulls.get(0);
-
- List<LocalAbstractObject> points = new ArrayList<>();
-
- // Minimal number of points in a hull is 3
- int maxCoveredPoints = -3;
- int maxIndex = hulls.get(0).hull.size();
-
- for (HullRepresentation hull : hulls) {
- points.addAll(hull.shared.getObjects());
- maxCoveredPoints += hull.hull.size();
- maxIndex = Math.max(maxIndex, hull.hull.size());
- }
-
- HashSet<LocalAbstractObject> coveredObjects = new HashSet<>();
- HullRepresentation newHull = new HullRepresentation(points);
-
- for (int i = 0; i < maxIndex; i++)
- for (HullRepresentation hull : hulls)
- if (i < hull.hull.size())
- checkCoveredObjects(hull, hulls, hull.hull.get(i), coveredObjects, maxCoveredPoints, newHull);
-
- hulls.forEach(hull -> addNotCoveredHullPoints(hull, newHull, coveredObjects));
-
- return newHull;
- }
-
- protected HullRepresentation mergeHulls(HullRepresentation first, HullRepresentation second){
- // create new hull
- List<LocalAbstractObject> data = new ArrayList<>(first.shared.getObjects());
- data.addAll(second.shared.getObjects());
- HullRepresentation newHull = new HullRepresentation(data);
-
- // prejdi vsetky body prveho a druheho a mergeni to dokopy
- HashSet<LocalAbstractObject> coveredObjects = new HashSet<>();
- sortHullPointsBasedOnMinDistanceToOther(first, second);
- sortHullPointsBasedOnMinDistanceToOther(second, first);
-// printPoints(first, second, path, hullNum, idx);
- int maxCoveredPointsNumber = first.hull.size() + second.hull.size() - 3; //minimalne chceme ponechat 3 hullPoints
- int maxIVal = (first.hull.size() > second.hull.size()) ? first.hull.size() : second.hull.size(); // max index to check
-
- for (int i = 0; i < maxIVal; i++) {// check iteratively both hulls whether their points are covered
- if (first.hull.size() > i){
- checkCoveredObj(first, second, first.hull.get(i), coveredObjects, maxCoveredPointsNumber, newHull);
- }
- if (second.hull.size() > i){
- checkCoveredObj(first, second, second.hull.get(i), coveredObjects, maxCoveredPointsNumber, newHull);
- }
- }
- addNotCoveredHullPoints(first, newHull, coveredObjects);
- addNotCoveredHullPoints(second, newHull, coveredObjects);
- return newHull;
- }
-
- private void sortHullPointsBasedOnMinDistanceToOther(HullRepresentation first, HullRepresentation second){
- List<LocalAbstractObject> sorted = new ArrayList<>();
- int[] hullIdxs = new int[first.hull.size()];
- float[] minDists = new float[first.getRepresentativesCount()];
- for (int i = 0; i < first.hull.size(); i++) {
- minDists[i] = this.shared.minDist(this.shared.getIndex(first.hull.get(i)),
- this.shared.getIndexes(second.hull)); // vzdialenost pocitam nad vsetkymi precomputed distances,
- hullIdxs[i] = i; // creating hullIndexes
- }
- Ranked[] ranked = getRankedIdxs(hullIdxs, minDists); // chcem zoradit firstHullIndexes
-
- for (int i = 0; i < first.getRepresentativesCount(); i++)
- sorted.add(first.hull.get(ranked[i].index));
- first.hull = sorted;
- }
-
- /** TODO: Vlasta 2020/06/23 -- should not we try to replace an existing hull object in newHull instead of "blindly" adding the hull object? */
- private void addNotCoveredHullPoints(HullRepresentation hull, HullRepresentation newHull,
- HashSet<LocalAbstractObject> coveredObjects){
- for (LocalAbstractObject hullPoint : hull.hull) {
- if (coveredObjects.contains(hullPoint))
- continue;
- newHull.hull.add(hullPoint);
- }
- }
-
- private void checkCoveredObjects(HullRepresentation firstHull,
- List<HullRepresentation> hulls,
- LocalAbstractObject excludedPoint,
- HashSet<LocalAbstractObject> coveredPoints,
- int maxCoveredPoints,
- HullRepresentation newHull) {
- List<HullRepresentation> everyHull = new ArrayList<>();
- everyHull.add(firstHull);
- everyHull.addAll(hulls);
-
- List<LocalAbstractObject> pointsForCoverageCheck = getPointsToComputeCoverage(everyHull, excludedPoint, coveredPoints);
- if (isCoveredByHull(shared.getIndex(excludedPoint), pointsForCoverageCheck.size(), this.shared.getIndexes(pointsForCoverageCheck)) && coveredPoints.size() < maxCoveredPoints) {
- if (this.mergingType == MergingType.BASIC) {
- checkCoverageWithOrWithoutHullPoint(newHull, excludedPoint, coveredPoints, pointsForCoverageCheck);
- return;
- }
-
- coveredPoints.add(excludedPoint);
- }
- }
-
- private void checkCoveredObj(HullRepresentation first, HullRepresentation second, LocalAbstractObject hullPoint,
- HashSet<LocalAbstractObject> coveredObjects, int maxCoveredPointsNumber, HullRepresentation newHull){
- List<LocalAbstractObject> pointsForCoverageCheck =
- getPointsToComputeCoverage(first, second, hullPoint, coveredObjects);
- if (isCoveredByHull(shared.getIndex(hullPoint), pointsForCoverageCheck.size(),
- this.shared.getIndexes(pointsForCoverageCheck)) &&
- coveredObjects.size() < maxCoveredPointsNumber ){
- if (this.mergingType == MergingType.BASIC
- // && hullArray.size() <= Math.sqrt(this.shared.getObjectCount())
- ) {
- checkCoverageWithOrWithoutHullPoint(newHull, hullPoint, coveredObjects, pointsForCoverageCheck);
- return;
- }
- coveredObjects.add(hullPoint); // just simply add the hull points to covered
- }
- }
-
- private void checkCoverageWithOrWithoutHullPoint(HullRepresentation newHull, LocalAbstractObject hullPoint,
- HashSet<LocalAbstractObject> coveredObjects,
- List<LocalAbstractObject> pointsForCoverageCheck) {
- // check whether the coverage is better when we consider the hullPoint to be "left" between the hull points or on,
- newHull.hull = pointsForCoverageCheck; // ptsFCCh contains only so far valid candidates for remaining hull pts
- newHull.setCoveredByHull();
- float coverageWithoutHP = newHull.getCoverage();
- newHull.hull.add(hullPoint); // try to add new point
- newHull.setCoveredByHull();
- float coverageWithHP = newHull.getCoverage();
- newHull.hull = new ArrayList<>(); //reset
- if (coverageWithoutHP >= coverageWithHP) {
- coveredObjects.add(hullPoint);
- } // else ignore
-
- }
-
- private void printPoints(HullRepresentation first, HullRepresentation second, String path, int i, int idx) {
- try {
- System.out.println("hulls");
- first.write(System.out);
- second.write(System.out);
- OutputStream dataS;
- if(idx != -1){
- // rewrite the data ... if we are adding the third hull
- dataS = new FileOutputStream(path + "d" + idx + ".txt", false);
- }
- else {
- dataS = new FileOutputStream(path + "d" + i + ".txt", false);
- }
- first.writeData(dataS);
- second.writeData(dataS);
-
- } catch (IOException e) {
- e.printStackTrace();
- }
- }
-}
\ No newline at end of file
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/HullIncrementalRepresentationV2.java b/src/cz/muni/fi/disa/similarityoperators/cover/HullIncrementalRepresentationV2.java
deleted file mode 100644
index 2b9afe1..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/HullIncrementalRepresentationV2.java
+++ /dev/null
@@ -1,400 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import messif.objects.LocalAbstractObject;
-import mhtree.HullNode;
-import mhtree.HullTree;
-
-import java.io.FileOutputStream;
-import java.io.IOException;
-import java.io.OutputStream;
-import java.util.*;
-
-/** TODO: This looks like it should not inherit from HullIncrementalRepresentation (almost all is overwritten) or
- * it must be refactored (creates its own hulls array?).
- *
- * Incrementally build a hull over the passed dataset -- version 2.
- *
- * Initial 3-object hulls are created from farthest object and their two neighbors. Their 3rd "neighbor" is selected as having min sum to 1st and 2nd.
- * Iteratively the initial hulls are merged -- the closest hulls first. The closest hulls HS to a hull H is obtained on hull objects, i.e.
- * the closest objects win (i.e. 1NN join(H,HS) is executed).
- * The iterations are "generation"-based -- all hulls are merged first, then such hulls are taken to the next iteration. The reduction in the number of hulls is 2.
- *
- * @author Miriama Janosova, Masaryk University, Brno, Czech Republic, x424615@fi.muni.cz
- */
-public class HullIncrementalRepresentationV2 extends HullIncrementalRepresentation {
-
- // MH-Tree
- private HullTree mhtree;
- private final HullNode[] nodes;
- private int hullsMergedIntoGroupsOf = 3;
-
- private int initialHullsSize = 3;
-
- private HullRepresentation[] hulls;
- private BitSet validHullIdxs;
- private boolean joinWithNotProcessedClosest;
-
- public HullIncrementalRepresentationV2(List<LocalAbstractObject> data, boolean joinWithNotProcessedClosest) {
- super(data);
-
- // MH-Tree
- nodes = new HullNode[shared.data.size() / initialHullsSize];
-
- hulls = new HullRepresentation[data.size() / initialHullsSize];
- validHullIdxs = new BitSet(hulls.length);
- validHullIdxs.set(0, hulls.length); // each index is valid,
- hullMinDistances = new float[hulls.length][hulls.length];
- this.joinWithNotProcessedClosest = joinWithNotProcessedClosest;
- }
-
- public HullIncrementalRepresentationV2(List<LocalAbstractObject> data, boolean joinWithNotProcessedClosest, int initialHullsSize) {
- super(data);
-
- this.initialHullsSize = Math.max(initialHullsSize, 3);
-
- // MH-Tree
- nodes = new HullNode[shared.data.size() / this.initialHullsSize];
-
- hulls = new HullRepresentation[data.size() / this.initialHullsSize];
- validHullIdxs = new BitSet(hulls.length);
- validHullIdxs.set(0, hulls.length); // each index is valid,
- hullMinDistances = new float[hulls.length][hulls.length];
- this.joinWithNotProcessedClosest = joinWithNotProcessedClosest;
- }
-
- public HullIncrementalRepresentationV2(PrecomputedDistances shared, boolean joinWithNotProcessedClosest) {
- super(shared);
-
- // MH-Tree
- nodes = new HullNode[shared.data.size() / initialHullsSize];
-
- hulls = new HullRepresentation[shared.data.size() / initialHullsSize];
- validHullIdxs = new BitSet(hulls.length);
- validHullIdxs.set(0, hulls.length); // each index is valid,
- hullMinDistances = new float[hulls.length][hulls.length];
- this.joinWithNotProcessedClosest = joinWithNotProcessedClosest;
- }
-
- public HullIncrementalRepresentationV2(List<LocalAbstractObject> data) {
- this(data, true);
- }
-
- public HullIncrementalRepresentationV2(PrecomputedDistances shared) {
- this(shared, true);
- }
-
- @Override
- protected boolean initHullPoints() {
- if (shared.data.size() < initialHullsSize) {
- hull.addAll(shared.data);
-
- // MH-Tree
- HullRepresentation newHull = new HullRepresentation(hull);
- newHull.hull.addAll(hull);
- mhtree = new HullTree(new HullNode(newHull));
-
- return false;
- }
-
- BitSet notProcessedIndices = new BitSet(shared.getObjectCount());
- notProcessedIndices.set(0, shared.getObjectCount());
- int hullIndex = 0;
-
- while (!notProcessedIndices.isEmpty()) {
- if (notProcessedIndices.cardinality() < initialHullsSize) {
- for (int i = notProcessedIndices.nextSetBit(0); i >= 0; i = notProcessedIndices.nextSetBit(i + 1))
- addObjectToClosestHull(i);
-
- return true;
- }
-
- List<LocalAbstractObject> hullPoints = new ArrayList<>();
-
- // Select a base point
- int furthestIndex = getFurthestIdx(notProcessedIndices);
- notProcessedIndices.clear(furthestIndex);
- hullPoints.add(shared.getObject(furthestIndex));
-
- // Select the rest of the `initialHullsSize` number of points
- findClosestIndices(furthestIndex, initialHullsSize - 1, notProcessedIndices)
- .forEach(i -> hullPoints.add(shared.getObject(i)));
-
- HullRepresentation newHull = new HullRepresentation(hullPoints);
- newHull.hull.addAll(hullPoints);
-
- hulls[hullIndex] = newHull;
- nodes[hullIndex] = new HullNode(newHull);
-
- hullIndex++;
- }
-
- return true;
- }
-
- private List<Integer> findClosestIndices(int basePointIndex, int numberOfIndices, BitSet notProcessedIndices) {
- List<Integer> closestIndices = new ArrayList<>();
- List<Integer> hullPointIndices = new ArrayList<>();
-
- hullPointIndices.add(basePointIndex);
-
- for (int i = 0; i < numberOfIndices; i++) {
- List<Integer> candidateIndices = new ArrayList<>();
- List<Float> candidateDistances = new ArrayList<>();
-
- for (int index : hullPointIndices) {
- int nnIndex = getKNNIndexes(index, 1, notProcessedIndices)[0];
-
- float distanceSum = hullPointIndices.stream()
- .map(pointIndex -> shared.getDistance(pointIndex, nnIndex))
- .reduce(0f, Float::sum);
-
- candidateIndices.add(nnIndex);
- candidateDistances.add(distanceSum);
- }
-
- int smallestDistanceIndex = candidateDistances.indexOf(Collections.min(candidateDistances));
- int closestPointIndex = candidateIndices.get(smallestDistanceIndex);
-
- notProcessedIndices.clear(closestPointIndex);
-
- hullPointIndices.add(closestPointIndex);
- closestIndices.add(closestPointIndex);
- }
-
- return closestIndices;
- }
-
- /** TODO: This is the same as in HullIncrementalRepresentation but it operates on this.hulls and not on super.hullArray REFACTOR?!? */
- protected void addObjectToClosestHull(int index) {
- int[] closesIdx = getKNNIndexes(index, 1, getAllCurrentHullIndexes());
- // returns closest index within the data
- LocalAbstractObject object = shared.getObject(index);
- int closestIndex = getAllCurrentHullIndexes()[closesIdx[0]];
- HullRepresentation closestHull = hulls[findCorrespondingHullIdx(shared.getObject(closestIndex))];
- closestHull.shared.data.add(object); // add just to data
- if (isCoveredByHull(index, closestHull.getRepresentativesCount(), shared.getIndexes(closestHull.hull))) {
- return; // don't add to hulls
- }
- closestHull.hull.add(object); //otherwise add to hull points
- }
-
- @Override
- protected int findCorrespondingHullIdx(LocalAbstractObject object) {
- for (int i = 0; i < hulls.length; i++) {
- if (hulls[i].hull.contains(object))
- return i;
- }
- return -1; // some hull has to be closest, does not make sense otherwise
- }
-
- @Override
- protected List<LocalAbstractObject> getAllCurrentHullObjects() {
- List<LocalAbstractObject> hullList = new ArrayList<>();
- for (HullRepresentation h : hulls) {
- hullList.addAll(h.hull);
- }
- return hullList;
- }
-
- /** TODO: Refactor? It is the same as private float[][] precomputeHullDistances in super */
- @Override
- protected void precomputeHullDistances(){
- for (int i = 0; i < hulls.length; i++) { // pre vsetky hullsIdxs
- for (int j = i + 1; j < hulls.length; j++) { // prejdi zvysne
- float minDist = Float.MAX_VALUE;// udrzi min vzdialenost medzi vsetkymi komb hullPointov
- for (LocalAbstractObject firstPoint : hulls[i].shared.data){ // prejdi vsetky hull points prveho a najdi najmensiu vzial k hociakemu druhemu
- for (LocalAbstractObject secondPoint : hulls[j].shared.data){
- if (shared.getDistance(firstPoint, secondPoint) < minDist){
- minDist = shared.getDistance(firstPoint, secondPoint);
- }
- }
- }
- if (minDist == 0f)
- throw new RuntimeException("Zero distance between " + hulls[i].toString() +
- " and " + hulls[j].toString());
- hullMinDistances[i][j] = minDist;
- hullMinDistances[j][i] = minDist;
- }
- }
- }
-
- @Override
- public void select() {
- if (!initHullPoints()) return;
-
- precomputeHullDistances();
-
- while (validHullIdxs.cardinality() != 1) {
- BitSet notProcessedIndices = (BitSet) validHullIdxs.clone();
-
- while (!notProcessedIndices.isEmpty()) {
- if (notProcessedIndices.cardinality() < hullsMergedIntoGroupsOf) {
- List<Integer> restOfTheIndices = new ArrayList<>();
-
- for (int j = notProcessedIndices.nextSetBit(0); j >= 0; j = notProcessedIndices.nextSetBit(j + 1))
- restOfTheIndices.add(j);
-
- int baseIndex = restOfTheIndices.remove(0);
- mergeHulls(baseIndex, restOfTheIndices);
-
- break;
- }
-
- int furthestHullIndex = getFurthestIdx(hullMinDistances, notProcessedIndices);
- notProcessedIndices.clear(furthestHullIndex);
-
- List<Integer> nnIndices = new ArrayList<>();
-
- for (int i = 0; i < hullsMergedIntoGroupsOf - 1; i++) {
- int index = shared.minDistInArrayExceptIdx(hullMinDistances[furthestHullIndex], joinWithNotProcessedClosest ? notProcessedIndices : validHullIdxs, furthestHullIndex);
- notProcessedIndices.clear(index);
- nnIndices.add(index);
- }
-
- mergeHulls(furthestHullIndex, nnIndices);
- }
- }
-
- int rootIndex = validHullIdxs.nextSetBit(0);
-
- mhtree = new HullTree(nodes[rootIndex]);
- this.hull = hulls[rootIndex].hull;
-
- setCoveredByHull();
- }
-
- /** get closest hull within hulls */
- private int getClosetsHullIdx(int hullIndex){
- return shared.minDistInArrayExceptIdx(hullMinDistances[hullIndex], validHullIdxs, hullIndex);
- }
-
- private void mergeHulls(int baseHullIndex, List<Integer> indices) {
- if (indices.size() == 0) return;
-
- // Create list of all the hulls
- List<HullRepresentation> hulls = new ArrayList<>();
- hulls.add(this.hulls[baseHullIndex]);
- indices.forEach(i -> hulls.add(this.hulls[i]));
-
- // Create a new node
- HullRepresentation newHull = mergeHulls(hulls);
- HullNode newNode = new HullNode(newHull);
-
- // Set parent for every corresponding input node
- nodes[baseHullIndex].setParent(newNode);
- indices.forEach(i -> nodes[i].setParent(newNode));
-
- // Set children for the new node
- newNode.addChildren(nodes[baseHullIndex]);
- newNode.addChildren(indices.stream().map(i -> nodes[i]));
-
- // Cleanup the nodes array and place the new node onto the first index
- nodes[baseHullIndex] = newNode;
- indices.forEach(i -> nodes[i] = null);
-
- // Cleanup the merged hulls
- this.hulls[baseHullIndex] = newHull;
- indices.forEach(i -> {
- validHullIdxs.clear(i);
- this.hulls[i] = null;
- });
-
- updateHullDistances(baseHullIndex, indices);
- }
-
- private void mergeHullsByIndexes(int firstIdx, int secondIdx){
- // create new hull
- HullRepresentation first = hulls[firstIdx];
- HullRepresentation second = hulls[secondIdx];
- HullRepresentation newHull = mergeHulls(first, second);
-
- // MH-Tree
- HullNode parent = new HullNode(newHull);
-
- nodes[firstIdx].setParent(parent);
- nodes[secondIdx].setParent(parent);
-
- parent.addChildren(nodes[firstIdx], nodes[secondIdx]);
-
- nodes[firstIdx] = parent;
- nodes[secondIdx] = null;
-
-
- hulls[firstIdx] = newHull;
- validHullIdxs.clear(secondIdx); //flip second hull index, first hull is updated
- hulls[secondIdx] = null;
- updateHullDists(firstIdx, secondIdx);
- }
-
- private void updateHullDistances(int baseHullIndex, List<Integer> hullIndices) {
- if (hullIndices.size() == 0) return;
-
- for (int i = validHullIdxs.nextSetBit(0); i >= 0; i = validHullIdxs.nextSetBit(i + 1)) {
- float minDistance = hullMinDistances[baseHullIndex][i];
-
- for (int index : hullIndices)
- minDistance = Math.min(minDistance, hullMinDistances[index][i]);
-
- hullMinDistances[baseHullIndex][i] = minDistance;
- hullMinDistances[i][baseHullIndex] = minDistance;
- }
- }
-
- private void updateHullDists(int firstIdx, int secondIdx) {
- for (int i = validHullIdxs.nextSetBit(0); i >= 0; i = validHullIdxs.nextSetBit(i + 1)) {
- float minDist = (hullMinDistances[firstIdx][i] < hullMinDistances[secondIdx][i])
- ? hullMinDistances[firstIdx][i] : hullMinDistances[secondIdx][i];
- hullMinDistances[firstIdx][i] = minDist;
- hullMinDistances[i][firstIdx] = minDist;
- }
- }
-
- private List<LocalAbstractObject> getPointsToComputeCoverage(HullRepresentation first,
- HullRepresentation second,
- LocalAbstractObject notToIncludePoint,
- HashSet<LocalAbstractObject> covered){
- List<LocalAbstractObject> hullPoints = new ArrayList<>(first.hull);
- hullPoints.addAll(second.hull);
- hullPoints.remove(notToIncludePoint);
- hullPoints.removeAll(covered);
- return hullPoints;
- }
-
- private void printGeneration(String path) {
- try {
- int idx = 0;
- for (int i = validHullIdxs.nextSetBit(0); i >= 0;
- i = validHullIdxs.nextSetBit(i + 1)) {
- OutputStream dataS = new FileOutputStream(path + "d" + idx + ".txt", false);
- hulls[i].writeData(dataS); // write data
- System.out.println("new hull\n");
- OutputStream out;
- OutputStream ncOut;
-
- out = new FileOutputStream(path + "h" + idx + ".txt", false);
- ncOut = new FileOutputStream(path + "nc" + idx + ".txt", false);
-
- hulls[i].write(out); // write hp
- hulls[i].setCoveredByHull();
- hulls[i].printNotCovered(ncOut); // write not covered
- idx++;
- }
-
- } catch (IOException e) {
- e.printStackTrace();
- }
- }
-
- // MH-Tree
- public HullTree getMHTree() {
- return mhtree;
- }
-
- public HullIncrementalRepresentationV2 mergeHullsIntoGroupsOfSize(int groupSize) {
- if (mhtree == null)
- this.hullsMergedIntoGroupsOf = groupSize;
-
- return this;
- }
-}
-
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/HullIncrementalRepresentationV3.java b/src/cz/muni/fi/disa/similarityoperators/cover/HullIncrementalRepresentationV3.java
deleted file mode 100644
index 96fdd8b..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/HullIncrementalRepresentationV3.java
+++ /dev/null
@@ -1,27 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import messif.objects.LocalAbstractObject;
-
-import java.util.List;
-
-/**
- * This is just a wrapper to enable "agglomerative" way of merging hulls.
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class HullIncrementalRepresentationV3 extends HullIncrementalRepresentationV2 {
-
- public HullIncrementalRepresentationV3(List<LocalAbstractObject> data) {
- super(data, false);
- }
-
- public HullIncrementalRepresentationV3(PrecomputedDistances shared) {
- super(shared, false);
- }
-
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/HullKNHPRepresentation.java b/src/cz/muni/fi/disa/similarityoperators/cover/HullKNHPRepresentation.java
deleted file mode 100644
index ca8031e..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/HullKNHPRepresentation.java
+++ /dev/null
@@ -1,46 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import messif.objects.LocalAbstractObject;
-
-import java.util.List;
-
-/**
- * works above HullRepresentation with min hull points
- * checks whether each data point is covered by kNearest Hull points,
- * if not adds the point into hull.
- */
-public class HullKNHPRepresentation extends HullMHPRepresentation{
-
- private int kNearestHullPoints;
-
- public HullKNHPRepresentation(List<LocalAbstractObject> data, int minHullPoints, int kNearestHullPoints) {
- super(data, minHullPoints);
- this.kNearestHullPoints = kNearestHullPoints;
- }
-
- public HullKNHPRepresentation(PrecomputedDistances shared, int minHullPoints, int kNearestHullPoints) {
- super(shared, minHullPoints);
- this.kNearestHullPoints = kNearestHullPoints;
- }
-
- @Override
- protected void setCoveredByHull() {
- setCoveredByKNHullPoints(kNearestHullPoints);
- }
-
- private void setCoveredByKNHullPoints(int knHullPoints) {
- int[] hullIdxs = shared.getIndexes(hull);
- setCoveredByHullShared(hullIdxs);
-
- for (int i = 0; i < shared.getObjectCount(); i++) {
- if (isCovered(i)) // Skip hull objects that were set above
- continue;
- if (isCoveredByHull(i, knHullPoints, hullIdxs)) // checks for coverage by only kn hull points
- setCovered(i);
- }
- }
-
- public int getkNearestHullPoints() {
- return kNearestHullPoints;
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/HullMHPRepresentation.java b/src/cz/muni/fi/disa/similarityoperators/cover/HullMHPRepresentation.java
deleted file mode 100644
index e3ddfda..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/HullMHPRepresentation.java
+++ /dev/null
@@ -1,43 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import messif.objects.LocalAbstractObject;
-
-import java.util.List;
-
-public class HullMHPRepresentation extends HullOptimizedRepresentationV2{
-
- private int minHullPoints;
-
- public HullMHPRepresentation(List<LocalAbstractObject> shared, int minHullPoints){
- super(shared);
- this.minHullPoints = minHullPoints;
- }
-
- public HullMHPRepresentation(PrecomputedDistances shared, int minHullPoints) {
- super(shared);
- this.minHullPoints = minHullPoints;
- }
-
- @Override
- protected boolean initHullPoints(){
- // not enough data to proceed
- if (shared.data.size() < minHullPoints){
- hull.addAll(shared.data);
- return false;
- }
- hull.add(getFurthest()); //get first hull point
- // add other hull points
- for (int i = 1; i < minHullPoints; i++){
- hull.add(nextFurthest());
- }
-
- hull.forEach((ho) -> presenceInHull.put(ho, 1));
-
- setCoveredByHull();
- return true;
- }
-
- public int getMinHullPoints() {
- return minHullPoints;
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/HullOptimizedRepresentation.java b/src/cz/muni/fi/disa/similarityoperators/cover/HullOptimizedRepresentation.java
deleted file mode 100644
index 32353d9..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/HullOptimizedRepresentation.java
+++ /dev/null
@@ -1,104 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import messif.objects.LocalAbstractObject;
-
-import java.util.ArrayList;
-import java.util.List;
-
-
-/**
- * Version 2 of the proposal for metric hull objects.
- *
- * It starts with two mutually furthest objects and one additional selected as an object having
- * the maximum sum of distance to the previous hull objects and is not close to any previous hull object.
- *
- * Until the data-set is covered (completely) repeat:
- * - select a new hull object in the same way as the third point above.
- * - compute coverage as if the next hull object was added.
- * - try to replace any existing hull object with the new one (so without adding a new object to the hull):
- * - compute coverage. If it is higher than the coverage of the hull extended with the new object,
- * choose the maximum coverage improvement and replace that object with the new one.
- * - if there have not been any improvement (replacement), add the new object to the hull.
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class HullOptimizedRepresentation extends HullRepresentation {
- //private int bannedSize;
- //private Set<AbstractObjectKey> bannedKeys;
-
- public HullOptimizedRepresentation(List<LocalAbstractObject> data) {
- super(data);
- //bannedSize = 0;
- //bannedKeys = new HashSet<>();
- }
-
- public HullOptimizedRepresentation(PrecomputedDistances shared) {
- super(shared);
- //bannedSize = 0;
- //bannedKeys = new HashSet<>();
- }
-
- @Override
- public void select() {
- if (!initHullPoints()){
- return;
- }
- while (!isAllCovered()) {
- List<LocalAbstractObject> notCovered = getNotCovered();
- // Next candidate
- hull.add(nextFurthest());
- setCoveredByHull();
- notCovered = getNotCovered();
-
- System.out.print("# Selected next hull object: "); System.out.println(toString());
-
- optimize();
- }
- }
-
- /*@Override
- protected boolean filter(int objIndex, LocalAbstractObject obj) {
- return bannedKeys.contains(obj.getObjectKey());
- }*/
-
- protected void optimize() {
- LocalAbstractObject hoNew = hull.get(hull.size() - 1); // new object added to the hull
- float coverageAll = getCoverage();
- float coverageBest = coverageAll;
-
- int indexBest = -1;
- List<LocalAbstractObject> notCovered = new ArrayList<LocalAbstractObject>();
-
- for (int i = hull.size() - 2; i >= 0; i--) {
- LocalAbstractObject hoRemoved = hull.remove(i);
- setCoveredByHull();
-
- notCovered = getNotCovered();
-
- float cvg = getCoverage();
- if (cvg > coverageBest) {
- coverageBest = cvg;
- indexBest = i;
- }
- hull.add(i, hoRemoved);
- }
- // Improved by replacement, so do it
- if (indexBest >= 0) {
- LocalAbstractObject hoRemoved = hull.remove(indexBest);
- // Add the object to the banned list
- //bannedKeys.add(hoRemoved.getObjectKey());
- } else {
- // Enlarged, so clear the banned list
- //bannedKeys.clear();
- //bannedSize = hull.size();
- }
- // Update the coverage back to original state!
- setCoveredByHull();
- notCovered = getNotCovered();
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/HullOptimizedRepresentationV2.java b/src/cz/muni/fi/disa/similarityoperators/cover/HullOptimizedRepresentationV2.java
deleted file mode 100644
index 8d263b0..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/HullOptimizedRepresentationV2.java
+++ /dev/null
@@ -1,167 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import messif.objects.LocalAbstractObject;
-
-import java.util.*;
-
-/**
- * Initially selects three mutually furthest objects.
- * Next hull object is iteratively selected out of remaining non-covered objects that has the best coverage.
- * If the resulting coverage is worse than before adding it, optimization is executed.
- * The optimization tries to expell any previous hull object.
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class HullOptimizedRepresentationV2 extends HullRepresentation {
- /** Counts how many times an object (key) was present/selected as a hull object */
- protected Map<LocalAbstractObject, Integer> presenceInHull = new HashMap<>();
-
- public HullOptimizedRepresentationV2(List<LocalAbstractObject> data) {
- super(data);
- }
-
- public HullOptimizedRepresentationV2(PrecomputedDistances shared) {
- super(shared);
- }
-
- protected boolean initHullPoints(){
- if (shared.data.size() < 3){
- hull.addAll(shared.data);
- return false;
- }
-
- // Start from the outlier
- hull.add(getFurthest());
- hull.add(nextFurthest());
- hull.add(nextFurthest());
- hull.forEach((ho) -> presenceInHull.put(ho, 1));
-
- setCoveredByHull();
- return true;
- }
-
- @Override
- public void select() {
- if (!initHullPoints())
- return;
-
- float prevCov = getCoverage();
- while (!isAllCovered()) {
- float cov = addNextHullObject();
- System.err.print("# Selected next hull object: "); System.err.println(toString());
- if (cov < prevCov) {
- // Optimization by replacement
- prevCov = optimize();
- } else {
- prevCov = cov;
- }
- reportNonCovered();
- }
- }
-
- /**
- *
- * @return resulting coverage after adding new hull object
- */
- protected float addNextHullObject() {
- float bestCoverage = -1;
- LocalAbstractObject bestObject = null;
- int[] hullIndexes = shared.getIndexes(hull);
- for (int i = 0; i < shared.getObjectCount(); i++) {
- if (isCovered(i) || inHull(i, hullIndexes))
- continue;
-
- LocalAbstractObject o = shared.getObject(i);
- hull.add(o);
- setCoveredByHull();
- System.err.println("# Testing next hull object: " + o.getLocatorURI());
- reportNonCovered();
- float coverage = getCoverage();
- if (coverage > bestCoverage) { // toto sposobuje problemy s cyklenim
- bestCoverage = coverage;
- bestObject = o;
- } else if (coverage == bestCoverage) {
- // Prefer object not selected yet
- if (presenceInHull.getOrDefault(o, 0) < presenceInHull.getOrDefault(bestObject, 0)) {
- bestObject = o;
- }
- }
- hull.remove(hull.size()-1);
- setCoveredByHull();
- }
- if (bestObject == null) {
- System.out.println("No object has been added.");
- return bestCoverage; // can this happen ?
- }
- // Add the best object and update internal structure about covered objects
- hull.add(bestObject);
- setCoveredByHull();
- presenceInHull.put(bestObject, presenceInHull.getOrDefault(bestObject, 0) + 1);
- return bestCoverage;
- }
-
- /**
- *
- * @return resulting coverage after optimization
- */
- protected float optimize() {
- float coverageBest = getCoverage(); // initial coverage
- int indexToDel = -1;
- LocalAbstractObject hoToDel = null;
-
- int medianPresence = getMedianPresence();
-
- //for (int i = hull.size() - 2; i >= 0; i--) {
- for (int i = 0; i < hull.size() - 1; i++) {
- LocalAbstractObject hoRemoved = hull.remove(i);
- setCoveredByHull();
- float cvg = getCoverage();
- if (cvg > coverageBest && presenceInHull.getOrDefault(hoRemoved, 0) <= medianPresence) {
- coverageBest = cvg;
- indexToDel = i;
- hoToDel = hoRemoved;
- } else if (cvg == coverageBest) {
- // Prefer object present in the hull fewer times
- if (presenceInHull.getOrDefault(hoRemoved, 0) < presenceInHull.getOrDefault(hoToDel, 0)) { // Vlasta, 2020/05/26 -- should not it be >???
- indexToDel = i;
- hoToDel = hoRemoved;
- }
- }
- hull.add(i, hoRemoved);
- }
- // Improved by replacement, so apply the removal
- if (indexToDel >= 0) {
- hull.remove(indexToDel);
- // Update the coverage back to original state!
- setCoveredByHull();
- System.err.print("# Removed hull object " + hoToDel.getLocatorURI() + " at " + indexToDel + " to increase coverage: "); System.err.println(toString());
- } else {
- setCoveredByHull(); // Restore original state
- }
- return coverageBest;
- }
-
- @Override
- public String toString() {
- String base = super.toString();
- StringBuilder sb = new StringBuilder(base);
- sb.append('\n').append("Presence in hull:");
- for (Map.Entry<LocalAbstractObject, Integer> e : presenceInHull.entrySet()) {
- String key = e.getKey().getLocatorURI();
- Integer value = e.getValue();
- sb.append(' ').append(key).append(':').append(value);
- }
- return sb.toString();
- }
-
- private int getMedianPresence() {
- List<Integer> values = new ArrayList<>(presenceInHull.values());
- Collections.sort(values);
- return values.get((int)(values.size() / 2));
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/HullOptimizedRepresentationV3.java b/src/cz/muni/fi/disa/similarityoperators/cover/HullOptimizedRepresentationV3.java
deleted file mode 100644
index b78bfb8..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/HullOptimizedRepresentationV3.java
+++ /dev/null
@@ -1,108 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import messif.objects.LocalAbstractObject;
-
-import java.util.ArrayList;
-import java.util.List;
-
-/**
- * Testing hull optimized representation, which may overcome the infinite loops which are a problem of the optimized and optimizedV2 hull representations.
- *
- * The technique is based on creating hashes for each size of the hull (i.e. we store the hashes only for hulls of the same cardinality),
- * such that if we end up in a situation where the current hull hashes to an already computed hash,
- * we consider this situation as loop, and therefore in the next step ??? we do not optimize(). The hull size would therefore necessarily increase despite possible coverage loss.
- * In the end the most desirable hull object would be selected (i.e. replacement of the objects takes place in the optimize() method).
- *
- * The implementation guarantees that after each iteration of select() the size of the hull either remains unchanged or increases.
- */
-public class HullOptimizedRepresentationV3 extends HullOptimizedRepresentationV2 {
-
- List<Integer> currentHullSizeHashes;
-
- public HullOptimizedRepresentationV3(List<LocalAbstractObject> data) {
- super(data);
- currentHullSizeHashes = new ArrayList<>();
- }
-
- public HullOptimizedRepresentationV3(PrecomputedDistances shared) {
- super(shared);
- currentHullSizeHashes = new ArrayList<>();
- }
-
- protected boolean initHullPoints(){
- if (shared.data.size() < 3){
- hull.addAll(shared.data);
- return false;
- }
-
- // Start from the outlier
- hull.add(getFurthest());
- hull.add(nextFurthest());
- hull.add(nextFurthest());
-
- setCoveredByHull();
- currentHullSizeHashes.add(hashCode());
-
- return true;
- }
-
- @Override
- public void select() {
- if (!initHullPoints())
- return;
-
- float prevCov = getCoverage();
- while (!isAllCovered()) {
- // po pridani objektu sa nutne zvacsi pocet objektov, ale nemusi to takto zostat nastalo.. kvoli metode optimize
- float cov = addNextHullObject();
- int hullObjectsCountBeforeOptimization = hull.size();
- System.out.print("# Selected next hull object: "); System.out.println(toString());
- if (cov < prevCov) {
- // Optimization by replacement
- prevCov = optimize();
- //newly added hull object has replaced some 'older' one, therefore the size of the hull stays the same
- if (hullObjectsCountBeforeOptimization == hull.size()){
- // nedoslo k optimalizacii, teda nic nemaz :)
- resetCurrentHullHashesList();
- continue;
- }
- int currentHash = hashCode();
- if (currentHullSizeHashes.contains(currentHash)){
- // the current hull object has already been selected before, therefore this may cause infinite loop.
- // we have to add another hull object, so that the size of hull necessarily increases.
- // afterwards another optimization steps may occur, which would lead to better selection of hull objects
- addNextHullObject();
- System.out.print("# Selected next hull object: ");
- System.out.println(toString());
- resetCurrentHullHashesList();
- }
- else {
- currentHullSizeHashes.add(currentHash);
- }
- } else {
- prevCov = cov;
- // no optimizations, therefore the size of hull increased, hence we have to delete the currentHullSizeHashes
- resetCurrentHullHashesList();
- }
- }
- }
-
- protected void resetCurrentHullHashesList(){
- currentHullSizeHashes = new ArrayList<>();
- currentHullSizeHashes.add(hashCode());
- }
-
-
- /**
- Computes hash of the currently selected hullPoints
- */
- public int hashCode() {
- int hash = 7;
- int prime = 31;
- for (LocalAbstractObject o : hull){
- hash = prime * hash + o.dataHashCode();
- }
- return hash;
- }
-
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/HullPreselectedRepresentation.java b/src/cz/muni/fi/disa/similarityoperators/cover/HullPreselectedRepresentation.java
deleted file mode 100644
index e5431f1..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/HullPreselectedRepresentation.java
+++ /dev/null
@@ -1,66 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import messif.objects.LocalAbstractObject;
-
-import java.util.List;
-
-/**
- * The input file (passed to {@link #select} method) is assumed to be a list of hull objects.
- *
- * This hull representation adds more and more hull objects and tries to find the best coverage.
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class HullPreselectedRepresentation extends AbstractHullRepresentation<HullPreselectedRepresentation> {
- public static final int MAX_HULL_SIZE = 150;
-
- public HullPreselectedRepresentation(List<LocalAbstractObject> data) {
- super(data);
- }
-
- public HullPreselectedRepresentation(PrecomputedDistances shared) {
- super(shared);
- }
- @Override
- protected void select() {
- // Take the first 3
- hull.add(shared.getObject(0));
- hull.add(shared.getObject(1));
- hull.add(shared.getObject(2));
-
- setCoveredByHull();
-// System.err.println("Initial hull: " + hull.size() + " objects");
-// System.err.println("Coverage: " + (int)Math.round(getCoverage() * 100f) + "%");
-// System.err.println("Cov ratio: " + getCoverageScore());
-
- float bestScore = getCoverageScore();
- int bestIndex = 2;
-
- for (int i = 3; !isAllCovered() && i < shared.getObjectCount() && i < MAX_HULL_SIZE; i++) {
- LocalAbstractObject o = shared.getObject(i);
- hull.add(o);
- setCoveredByHull();
- float score = getCoverageScore();
- if (score >= bestScore) {
- bestScore = score;
- bestIndex = i;
-// System.err.println("Preliminary hull: " + hull.size() + " objects");
-// System.err.println("Coverage: " + (int)Math.round(getCoverage() * 100f) + "%");
-// System.err.println("Cov ratio: " + getCoverageScore());
- }
- }
- // Remove all hull objects until we get the best configuration.
- while ((hull.size() - 1) > bestIndex)
- hull.remove(hull.size() - 1);
-
- setCoveredByHull();
- System.err.println("Preselected hull: " + hull.size() + " objects");
- System.err.println("Coverage: " + (int)Math.round(getCoverage() * 100f) + "%");
- System.err.println("Cov ratio: " + getCoverageScore());
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/HullRepresentation.java b/src/cz/muni/fi/disa/similarityoperators/cover/HullRepresentation.java
deleted file mode 100644
index 1055551..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/HullRepresentation.java
+++ /dev/null
@@ -1,139 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import messif.objects.LocalAbstractObject;
-import messif.utility.FileUtils;
-
-import java.io.IOException;
-import java.util.ArrayList;
-import java.util.List;
-
-/**
- * Original proposal for metric hull objects.
- *
- * It starts with two mutually furthest objects and one additional selected as an object having
- * the maximum sum of distance to the previous hull objects and is not close to any previous hull object.
- *
- * Until the data-set is covered (completely) a new hull object is appended. It is selected in the same way as
- * the third point above.
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class HullRepresentation extends AbstractHullRepresentation<HullRepresentation> {
-
- /** Instantiate a hull representation from a serialized file containing hull objects.
- * @param locatorID ID/symbolic name of this representation
- * @param objCls class of objects
- * @param filename file containing MESSIF serialized objects
- * @throws IllegalArgumentException
- * @throws IOException
- */
- public HullRepresentation(String locatorID, Class<LocalAbstractObject> objCls, String filename) throws IllegalArgumentException, IOException {
- super(locatorID, FileUtils.readObjects(objCls, filename)); // We do not select hull objects here, we rather take existing ones (already selected).
- hull.addAll(getObjects());
- }
-
- public HullRepresentation(List<LocalAbstractObject> data) {
- super(data);
- }
-
- public HullRepresentation(PrecomputedDistances shared) {
- super(shared);
- }
-
- public void setHullObjects(List<LocalAbstractObject> hullObjects) {
- hull.addAll(hullObjects);
- }
-
- protected boolean initHullPoints(){
- if (shared.data.size() <= 3){
- hull.addAll(shared.data);
- return false;
- }
-
- hull.add(getFurthest());
- hull.add(nextFurthest());
- hull.add(nextFurthest());
-
- setCoveredByHull();
- return true;
- }
-
- @Override
- public void select() {
-
- if(!initHullPoints()){
- return;
- }
- while (!isAllCovered()) {
- // Next candidate
- hull.add(nextFurthest());
- setCoveredByHull();
- System.out.print("# Selected next hull object: "); System.out.println(toString());
- }
- }
-
- protected LocalAbstractObject nextFurthest() {
- float distThreshold = getMinDist(hull) / 2f;
- float maxSum = 0f;
- int maxIdx = -1;
- List<Integer> filtered = new ArrayList<>();
- boolean skipFilter = false;
- do {
- int[] hullIndexes = shared.getIndexes(hull);
- for (int i = 0; i < shared.getObjectCount(); i++) {
- if (isCovered(i) || inHull(i, hullIndexes))
- continue;
- LocalAbstractObject o = shared.getObject(i);
- if (!skipFilter && filter(i, o)) {
- filtered.add(i);
- continue;
- }
- float[] dists = shared.getDistances(i);
-
- float sum = 0;
- for (int hullIndex : hullIndexes) {
- float d = dists[hullIndex];
- if (d == 0 && (distThreshold/2 == 0f)){
- sum += distThreshold;
- continue;
- }
- if (d < distThreshold){
- sum = 0f;
- break; // object o is too close to an existing hull point, thus ignore
- }
- sum += d; // else add distance to hull point to sum
-// if (d > distThreshold)
-// sum += d;
- }
- if (sum > maxSum) {
- maxSum = sum;
- maxIdx = i;
- }
-
- }
- if (maxSum == 0 && !filtered.isEmpty()) { // (maxSum > 0) <=> we have checked at least one candidate.
- skipFilter = true;
- filtered.clear();
- }
- distThreshold /= 2f;
- } while (maxIdx == -1 && distThreshold > 0f);
-
- assert maxIdx != -1;
- return shared.getObject(maxIdx);
- }
-
- /**
- * Additional filter on selection of next furthest object to be added into the hull.
- * @param objIndex index to {@link #shared}
- * @param obj object at the objIndex
- * @return <code>true</code> to skip this object, <code>false</code> to include it in the candidates for the next furthest object.
- */
- protected boolean filter(int objIndex, LocalAbstractObject obj) {
- return false; // We do not any additional filtering here
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/MinDistObjectPairIterator.java b/src/cz/muni/fi/disa/similarityoperators/cover/MinDistObjectPairIterator.java
deleted file mode 100644
index f062dcb..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/MinDistObjectPairIterator.java
+++ /dev/null
@@ -1,89 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import cz.muni.fi.disa.similarityoperators.cover.AbstractRepresentation.MinDistObjectPair;
-
-import java.util.BitSet;
-import java.util.Collections;
-import java.util.Iterator;
-import java.util.PriorityQueue;
-
-/**
- * Iterates over all pairs of objects (using the pair-wise distance matrix) and returns pairs of objects having the smallest distance.
- * It skips already processed objects, so they cannot be reported once marked in the passed BitSet (its bit is cleared).
- * Caching is implemented to improve performance.
- *
- * @author Miriama Janosova, Masaryk University, Brno, Czech Republic, x424615@fi.muni.cz
- */
-public class MinDistObjectPairIterator implements Iterator<MinDistObjectPair> {
- private static final int CACHE_SIZE = 10;
-
- private float[][] distancesArray;
- private BitSet notProcessedIndexes; //stores which objects has been processed
- private MinDistObjectPair[] cache; //sorted array of minDistObjects
- private int nextMinDistIndex;
-
- public MinDistObjectPairIterator(float[][] distancesArray, BitSet notProcessedIdxs){
- this.distancesArray = distancesArray;
- this.notProcessedIndexes = notProcessedIdxs;
- this.cache = new MinDistObjectPair[CACHE_SIZE];
- this.nextMinDistIndex = 0;
- }
-
- @Override
- public boolean hasNext() {
- return notProcessedIndexes.cardinality() > 0;
- }
-
- @Override
- public MinDistObjectPair next() {
- while (true) {
- // either we are out of bounds or we do not have cache values anymore
- if (nextMinDistIndex >= CACHE_SIZE || cache[nextMinDistIndex] == null){
- recomputeMinDists();
- }
- MinDistObjectPair minDistsObject = cache[nextMinDistIndex]; //get current minDistObject
- cache[nextMinDistIndex] = null;
- nextMinDistIndex++;
- if(notProcessedIndexes.get(minDistsObject.firstIdx)
- && notProcessedIndexes.get(minDistsObject.secondIdx)){ // none of the indexes has been processed yet
- //therefore we flip them
- notProcessedIndexes.clear(minDistsObject.firstIdx);
- notProcessedIndexes.clear(minDistsObject.secondIdx);
- return minDistsObject; // we can return the object
- }
- //otherwise we continue
- }
- }
-
- private void recomputeMinDists() {
- PriorityQueue<MinDistObjectPair> maxHeap = new PriorityQueue<>(Collections.reverseOrder()); //max heap
- // we use max heap for storing CACHE_SIZE minimal distances between objects
- for (int i = notProcessedIndexes.nextSetBit(0); i != -1; i = notProcessedIndexes.nextSetBit(i + 1)) {
- for (int j = notProcessedIndexes.nextSetBit(i + 1); j != -1; j = notProcessedIndexes.nextSetBit(j + 1)) {
- float d = distancesArray[i][j];
- if (maxHeap.size() < CACHE_SIZE) {
- maxHeap.add(new MinDistObjectPair(i, j, d));
- continue;
- }
- if (d >= maxHeap.peek().distance) {
- continue; //dist greater than max_value in heap, therefore continue
- }
- maxHeap.poll(); //remove the old root
- maxHeap.add(new MinDistObjectPair(i, j, d)); // add new object
- }
- }
- transformMaxHeapIntoMinArray(maxHeap);
- }
-
- private void transformMaxHeapIntoMinArray(PriorityQueue<MinDistObjectPair> maxHeap) {
- int i = maxHeap.size(); // i can be smaller than CACHE_SIZE
- while (!maxHeap.isEmpty()) {
- cache[--i] = maxHeap.poll(); //extract the max value from heap, put it in the further possible position
- }
- nextMinDistIndex = 0; // reset of cache completed
- }
-
- public BitSet getNotProcessedIndexes(){
- return notProcessedIndexes;
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/SnakeAutoRepresentation.java b/src/cz/muni/fi/disa/similarityoperators/cover/SnakeAutoRepresentation.java
deleted file mode 100644
index d56ad26..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/SnakeAutoRepresentation.java
+++ /dev/null
@@ -1,382 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.cover;
-
-
-import cz.muni.fi.disa.similarityoperators.cover.types.BallBasedCover;
-import messif.objects.LocalAbstractObject;
-import messif.objects.keys.RadiusObjectKey;
-import messif.utility.FileUtils;
-
-import java.io.IOException;
-import java.util.*;
-import java.util.logging.Level;
-import java.util.logging.Logger;
-
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-
-/**
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class SnakeAutoRepresentation extends AbstractRepresentation<SnakeAutoRepresentation> implements BallBasedCover {
- private static final Logger logger = Logger.getLogger(SnakeAutoRepresentation.class.getSimpleName());
-
- private List<Ball> balls;
- private int nearestReverseNeighbors;
- private float stopDistance;
-
- private int[] reverseNeighborCount;
-
- /** Instantiate a snake representation from a serialized file containing objects one for a ball representation.
- * Each object must have the key of RadiusObjectKey that holds the ball radius.
- *
- * @param locatorID ID/symbolic name of this representation
- * @param objCls class of objects
- * @param filename file containing MESSIF serialized objects
- * @throws IllegalArgumentException
- * @throws IOException
- */
- public SnakeAutoRepresentation(String locatorID, Class<LocalAbstractObject> objCls, String filename) throws IllegalArgumentException, IOException {
- super(locatorID, Collections.EMPTY_LIST);
-
- this.nearestReverseNeighbors = 0;
- this.stopDistance = 0;
- this.reverseNeighborCount = new int[0];
-
- this.balls = new ArrayList<>();
- FileUtils.readObjects(objCls, filename).forEach((c) -> {
- balls.add(new Ball(c, c.getObjectKey(RadiusObjectKey.class).getRadius()));
- });
- }
-
- public SnakeAutoRepresentation(List<LocalAbstractObject> data, int nearestReverseNeighbors) {
- this(data, nearestReverseNeighbors, 0f);
- }
-
- public SnakeAutoRepresentation(PrecomputedDistances shared, int nearestReverseNeighbors) {
- this(shared, nearestReverseNeighbors, 0f);
- }
-
- public SnakeAutoRepresentation(List<LocalAbstractObject> data, int nearestReverseNeighbors, float stopDistance) {
- super(data);
- this.stopDistance = stopDistance;
- init(nearestReverseNeighbors);
- }
-
- public SnakeAutoRepresentation(PrecomputedDistances shared, int nearestReverseNeighbors, float stopDistance) {
- super(shared);
- this.stopDistance = stopDistance;
- init(nearestReverseNeighbors);
- }
-
- private void init(int nearestReverseNeighbors) {
- this.nearestReverseNeighbors = nearestReverseNeighbors;
- this.balls = new ArrayList<>();
- this.reverseNeighborCount = new int[shared.getObjectCount()];
- }
-
- @Override
- public String toString() {
- String base = toShortString();
- StringBuilder sb = new StringBuilder(base.substring(0, base.length()-1));
- sb.append(", ");
- boolean sep = false;
- for (Ball b : balls) {
- if (sep)
- sb.append("; ");
- sb.append(b);
- sep = true;
- }
- sb.append("]");
- return sb.toString();
- }
-
- @Override
- public String toShortString() {
- float min = Float.MAX_VALUE,
- max = 0f,
- sum = 0f;
- for (Ball b : balls) {
- if (b.getRadius() > max)
- max = b.getRadius();
- if (b.getRadius() < min)
- min = b.getRadius();
- sum += b.getRadius();
- }
- float avg = (balls.isEmpty()) ? 0f : (sum / (float)balls.size());
-
- StringBuilder sb = new StringBuilder(getClass().getSimpleName());
- sb.append('[');
- sb.append("objects=").append(shared.getObjectCount()).append(", ");
- sb.append("rnn=").append(nearestReverseNeighbors).append(", ");
- sb.append("stopDist=").append(stopDistance).append(", ");
- sb.append("centers=").append(balls.size()).append(", ");
- sb.append("radius_min=").append(min).append(", ");
- sb.append("radius_max=").append(max).append(", ");
- sb.append("radius_avg=").append(avg).append("]");
- return sb.toString();
- }
-
- @Override
- public Iterator<LocalAbstractObject> getRepresentatives() {
- return new Iterator<LocalAbstractObject>() {
- final Iterator<Ball> it = balls.iterator();
-
- @Override
- public boolean hasNext() {
- return it.hasNext();
- }
-
- @Override
- public LocalAbstractObject next() {
- return it.next().center;
- }
- };
- }
-
- @Override
- public int getRepresentativesCount() {
- return balls.size();
- }
-
- @Override
- public float getAverageRadius() {
- float sum = 0f;
- for (Ball b : balls)
- sum += b.getRadius();
- return (sum == 0f) ? sum : (sum / balls.size());
- }
-
- @Override
- protected void select() {
- // Calculate counts of being "reverse nearest neighbor"
- for (int i = 0; i < shared.getObjectCount(); i++) {
- List<LocalAbstractObject> objs = getKNNObjects(shared.getObject(i), this.nearestReverseNeighbors);
- for (LocalAbstractObject o : objs)
- reverseNeighborCount[shared.getIndex(o)]++;
- }
- // Map number of being rkNN to object count
- Map<Integer,Integer> mapNNtoCnt = new TreeMap<>();
- // Get maximum
- int max = 0;
- for (int i = 0; i < reverseNeighborCount.length; i++) {
- int cnt = 1 + mapNNtoCnt.getOrDefault(reverseNeighborCount[i], 0);
- mapNNtoCnt.put(reverseNeighborCount[i], cnt);
- if (reverseNeighborCount[i] > max)
- max = reverseNeighborCount[i];
- }
- if (logger.isLoggable(Level.FINE)) {
- logger.log(Level.FINE, "Maximum of being {0}-RNN: {1}", new Object[]{nearestReverseNeighbors, max});
- StringBuilder sb = new StringBuilder();
- for (Map.Entry<Integer, Integer> e : mapNNtoCnt.entrySet())
- sb.append(e.getKey()).append('=').append(e.getValue()).append(' ');
- logger.log(Level.FINE, "Occurences of being {0}-RNN (rNN=obj_count): {1}", new Object[]{nearestReverseNeighbors, sb.toString()});
- }
-
- // Repeatedly add balls around the object having their RNN count equal to current max
- boolean repeat, repeat2 = false;
- int iterLimit = 20;
- int minCoverSize = 5; // can be proportional to the data-set size?
- do {
- if (--iterLimit == 0)
- break;
- // Select cover objects
- for (int i = 0; i < shared.getObjectCount(); i++) {
- if (reverseNeighborCount[i] == max && !isCovered(i)) {
- setCovered(i);
- balls.add(new Ball(shared.getObject(i), 0));
- }
- }
- // Set radii for cover objects
- if (balls.size() < minCoverSize) {
- // Next round automatically
- max -= 1;
- logger.log(Level.FINE, "Decrementing maximum of being {0}-RNN due to too few cover objects: {1}", new Object[]{nearestReverseNeighbors, max});
- repeat = true;
- continue;
- }
- computeBallRadii(false);
-
- List<BallPair> overlaps = new ArrayList<>();
- // Test if any ball is inside another cover ball
- Iterator<Ball> iter = balls.iterator();
- while (iter.hasNext()) {
- Ball b = iter.next();
- // Remove zero-radius balls automatically
- if (b.radius == 0f) {
- iter.remove();
- b.setRadius(-1f);
- logger.log(Level.FINE, "Removing zero-radius ball {0}", b);
- continue;
- }
- int bIdx = shared.getIndex(b.center);
- float[] bDists = shared.getDistances(bIdx);
- // Find a ball that contains 'b' completely
- for (Ball bOn : balls) {
- if (b == bOn) // Skip myself!
- continue;
- float d = bDists[shared.getIndex(bOn.center)];
- if (b.radius + d <= bOn.radius) {
- // bOn covers b completely
- iter.remove(); // do not clear its "covered" flag
- logger.log(Level.FINE, "Removing ball {0} because it is covered by ball {1}", new Object[]{b, bOn});
- b.setRadius(-1f); // a flag to recognize that this ball was removed!
- break;
- } else if (d <= Math.min(b.radius, bOn.radius)) {
- if (stopDistance >= Math.max(b.radius, bOn.radius))
- logger.log(Level.FINE, "Too small balls, so stoping: stopDists={0}, radius1={1}, radius2={2}",
- new Object[]{stopDistance, b.radius, bOn.radius});
- else
- overlaps.add(new BallPair(b, bOn, d));
- }
- }
- }
- // Test all pairs whether they are still valid
- repeat = (balls.size() < minCoverSize);
- repeat2 = false;
- for (BallPair bp : overlaps) {
- if (bp.left.getRadius() >= 0 || bp.right.getRadius() >= 0) {
- // OK, there is a major overlap, so decrement max.
- repeat2 = true;
- logger.log(Level.FINE, "Too close balls, removing the one covering fewer objects: "
- + "dist={0}, objectsI={1}, objectsJ={2}",
- new Object[]{bp.distance, bp.left.coveredObjects, bp.right.coveredObjects});
- if (bp.left.coveredObjects < bp.right.coveredObjects) {
- balls.remove(bp.left);
- bp.left.setRadius(-1);
- } else {
- balls.remove(bp.right);
- bp.right.setRadius(-1);
- }
- }
- }
- if (repeat2 || repeat) {
- max -= 1;
- logger.log(Level.FINE, "Decrementing maximum of being {0}-RNN: {1}", new Object[]{nearestReverseNeighbors, max});
- }
- } while (max > 0 && (repeat || repeat2));
-
- // Update radii and coverage flags for points inside
- computeBallRadii(true);
- }
-
- private void computeBallRadii(boolean setAsCovered) {
- if (balls.isEmpty())
- return;
- // Clear radii and covered counts first
- balls.forEach((b) -> { b.clear(); });
- // Set new radii
- for (int i = 0; i < shared.getObjectCount(); i++) {
- //if (isCovered(i)) // Assign all (ball centers can be processed since they are assigned to themselves)
- // continue;
-
- BallDistPair nearest = getNearestBall(shared.getObject(i));
- if (nearest.ball.getRadius() < nearest.distance)
- nearest.ball.setRadius(nearest.distance);
- nearest.ball.incrCoveredObjects();
- if (setAsCovered)
- setCovered(i);
- }
- }
-
-
- protected BallDistPair getNearestBall(LocalAbstractObject center) {
- int centerIdx = shared.getIndex(center);
- float[] dists = shared.getDistances(centerIdx);
-
- float distMin = Float.MAX_VALUE;
- Ball ballMin = null;
- for (Ball b : balls) {
- int ballIdx = shared.getIndex(b.center);
- if (dists[ballIdx] < distMin) {
- distMin = dists[ballIdx];
- ballMin = b;
- }
- }
- if (ballMin != null)
- return new BallDistPair(ballMin, distMin);
- else
- return null;
- }
-
-
- private List<LocalAbstractObject> getCenters() {
- List<LocalAbstractObject> lst = new ArrayList<>(balls.size());
- for (Ball b : balls)
- lst.add(b.center);
- return lst;
- }
-
- @Override
- public boolean isExternalCovered(LocalAbstractObject o) {
- for (Ball b : balls) {
- if (o.getDistance(b.center) <= b.radius)
- return true;
- }
- return false;
- }
-
- public static class Ball {
- private LocalAbstractObject center;
- private float radius;
- private int coveredObjects;
-
- public Ball(LocalAbstractObject center, float radius) {
- this.center = center;
- this.radius = radius;
- }
-
- public void clear() {
- setRadius(0f);
- coveredObjects = 0;
- }
-
- public void setRadius(float radius) {
- this.radius = radius;
- }
-
- public float getRadius() {
- return radius;
- }
-
- private void incrCoveredObjects() {
- coveredObjects++;
- }
-
- public int getCoveredObjects() {
- return coveredObjects;
- }
-
- @Override
- public String toString() {
- StringBuilder sb = new StringBuilder();
- sb.append("center=").append(center).append(", radius=").append(radius);
- return sb.toString();
- }
- }
-
- public static class BallDistPair {
- public final Ball ball;
- public final float distance;
-
- public BallDistPair(Ball ball, float distance) {
- this.ball = ball;
- this.distance = distance;
- }
- }
-
- public static class BallPair {
- public final Ball left;
- public final Ball right;
- public final float distance;
-
- public BallPair(Ball left, Ball right, float dist) {
- this.left = left;
- this.right = right;
- this.distance = dist;
- }
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/SnakeRepresentation.java b/src/cz/muni/fi/disa/similarityoperators/cover/SnakeRepresentation.java
deleted file mode 100644
index eae4a10..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/SnakeRepresentation.java
+++ /dev/null
@@ -1,134 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.cover;
-
-import messif.objects.LocalAbstractObject;
-import messif.objects.keys.RadiusObjectKey;
-import messif.utility.FileUtils;
-
-import java.io.IOException;
-import java.util.ArrayList;
-import java.util.Collections;
-import java.util.Iterator;
-import java.util.List;
-
-/**
- * Snake representation of data set.
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class SnakeRepresentation extends AbstractRepresentation<SnakeRepresentation> {
- private float maxRadius;
- private List<LocalAbstractObject> centers;
-
- /** Instantiate a snake representation from a serialized file containing objects one for a ball representation.
- * The first object must have the key of RadiusObjectKey that holds the ball radius.
- * The other objects can have this key too, but they are ignored.
- *
- * @param locatorID ID/symbolic name of this representation
- * @param objCls class of objects
- * @param filename file containing MESSIF serialized objects
- * @throws IllegalArgumentException
- * @throws IOException
- */
- public SnakeRepresentation(String locatorID, Class<LocalAbstractObject> objCls, String filename) throws IllegalArgumentException, IOException {
- super(locatorID, Collections.EMPTY_LIST);
- centers = FileUtils.readObjects(objCls, filename);
- maxRadius = centers.get(0).getObjectKey(RadiusObjectKey.class).getRadius();
- }
-
- public SnakeRepresentation(List<LocalAbstractObject> data, float maxRadius) {
- super(data);
- this.maxRadius = maxRadius;
- this.centers = new ArrayList<>();
- }
-
- public SnakeRepresentation(PrecomputedDistances shared, float maxRadius) {
- super(shared);
- this.maxRadius = maxRadius;
- this.centers = new ArrayList<>();
- }
-
- @Override
- public String toString() {
- StringBuilder sb = new StringBuilder(getClass().getSimpleName());
- sb.append('[');
- sb.append("objects=").append(shared.getObjectCount()).append(", ");
- sb.append("max_radius=").append(maxRadius).append(", ");
- sb.append("centers=").append(centers.size()).append(", ");
- boolean sep = false;
- for (LocalAbstractObject c : centers) {
- if (sep)
- sb.append("; ");
- sb.append("center=").append(c).append(", radius=").append(maxRadius);
- sep = true;
- }
- sb.append("]");
- return sb.toString();
- }
-
- @Override
- public String toShortString() {
- StringBuilder sb = new StringBuilder(getClass().getSimpleName());
- sb.append('[');
- sb.append("objects=").append(shared.getObjectCount()).append(", ");
- sb.append("max_radius=").append(maxRadius).append(", ");
- sb.append("centers=").append(centers.size()).append("]");
- return sb.toString();
- }
-
- @Override
- public Iterator<LocalAbstractObject> getRepresentatives() {
- return centers.iterator();
- }
-
- @Override
- public int getRepresentativesCount() {
- return centers.size();
- }
-
- @Override
- protected void select() {
- // Start from the mediod
- LocalAbstractObject cur = getMedoid();
- centers.add(cur);
- setCoveredByBall(cur, maxRadius);
-
- while (!isAllCovered()) {
- // Next candidate
- LocalAbstractObject next = closestCandidate();
- List<LocalAbstractObject> nextBall = getBallObjects(next, maxRadius, true);
-
- // Update candidate to mediod from object in the ball around "next"
- LocalAbstractObject med = getMedoid(nextBall);
- if (med != null)
- next = med;
-
- centers.add(next);
- setCoveredByBall(next, maxRadius);
- }
- }
-
- /** Next not-covered object that is closest to currently selected cover objects (ball centers) */
- private LocalAbstractObject closestCandidate() {
- float min = Float.MAX_VALUE;
- LocalAbstractObject cand = null;
-
- int[] centerIdxs = shared.getIndexes(centers);
-
- for (int i = 0; i < shared.getObjectCount(); i++) {
- if (isCovered(i))
- continue;
-
- float sum = shared.sumArray(i, centerIdxs);
- if (sum < min) {
- min = sum;
- cand = shared.getObject(i);
- }
- }
- return cand;
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/cover/types/BallBasedCover.java b/src/cz/muni/fi/disa/similarityoperators/cover/types/BallBasedCover.java
deleted file mode 100644
index 475e622..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/cover/types/BallBasedCover.java
+++ /dev/null
@@ -1,15 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.cover.types;
-
-/**
- * Marker for cover representatives based on metric balls.
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public interface BallBasedCover {
- float getAverageRadius();
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAll.java b/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAll.java
deleted file mode 100644
index c5b1b6a..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAll.java
+++ /dev/null
@@ -1,63 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.groupBy;
-
-import cz.muni.fi.disa.similarityoperators.cover.AbstractRepresentation.PrecomputedDistances;
-import cz.muni.fi.disa.similarityoperators.utils.OverlapType;
-import cz.muni.fi.disa.similarityoperators.utils.QueryType;
-import messif.objects.LocalAbstractObject;
-
-import java.util.*;
-
-// range-query ... an object belongs to the group if it is closer or at the same distance to every other object in the group
-// knn-query ...
-// important ! Unlike the SGB-Any, an object may belong to multiple groups in case the overlapping properties are met.
-// overlapping constraints : JOIN-ANY, ELIMINATE (discard object), FORM-NEW-GROUP (creates groups for overlapping objects .. todo nov group dokopy? pre kazdy zvlast?)
-public class SGBAll extends SGBQuery {
-
- public List<List<LocalAbstractObject>> executeQuery(PrecomputedDistances objects, QueryType queryType, double threshold, OverlapType overlapType) {
- List<Set<LocalAbstractObject>> groups = new ArrayList<>();
-
- for (LocalAbstractObject object : objects.getObjects()) {
- List<Integer> candidates = new ArrayList<>();
- List<Integer> overlap = new ArrayList<>();
- List<LocalAbstractObject> result = getSuitableObjects(object, objects, queryType, threshold);
-
- for (LocalAbstractObject o : result) {
- List<Integer> groupsIndexes = getGroupIdxsWithObject(o, groups);
- if (!groupsIndexes.isEmpty())
- overlap.addAll(groupsIndexes); // store group indexes
- }
- Set<Integer> distinctOverlap = new HashSet<>(overlap);
- for (Integer groupIndex : distinctOverlap) {
- // check whether each object in the group satisfies the threshold
- if (Collections.frequency(overlap, groupIndex) == groups.get(groupIndex).size())
- candidates.add(groupIndex);
- }
- if (candidates.isEmpty()) {
- Set<LocalAbstractObject> group = new HashSet<>();
- group.add(object);
- groups.add(group);
- }
- else if (candidates.size() == 1) {
- groups.get(candidates.get(0)).add(object);
- }
- else {
- switch (overlapType) {
- case JOIN_ANY: {
- groups.get(candidates.get(0)).add(object);
- break;
- }
- case FORM_NEW_GROUP: {
- Set<LocalAbstractObject> group = new HashSet<>();
- group.add(object);
- groups.add(group);
- break;
- }
- case ELIMINATE: {
- //todo .. toto je fakt bleh BitSet + ignoruj "zcernane prvky" ?
- }
- }
- }
- }
- return castInnerSetToList(groups);
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAll_N.java b/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAll_N.java
deleted file mode 100644
index c0b6910..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAll_N.java
+++ /dev/null
@@ -1,73 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.groupBy;
-
-import cz.muni.fi.disa.similarityoperators.cover.AbstractRepresentation;
-import cz.muni.fi.disa.similarityoperators.utils.ObjectGroups;
-import cz.muni.fi.disa.similarityoperators.utils.OverlapType;
-import cz.muni.fi.disa.similarityoperators.utils.QueryType;
-import messif.objects.LocalAbstractObject;
-
-import java.util.*;
-
-public class SGBAll_N extends SGBQuery {
-
- public ObjectGroups data; // object storing objects and computed clusters
-
- public SGBAll_N(AbstractRepresentation.PrecomputedDistances objects) {
- data = new ObjectGroups(objects);
- }
-
- public ObjectGroups executeQuery(QueryType queryType, double threshold, OverlapType overlapType) {
-
- for (int objIdx = 0; objIdx < data.objects.getObjectCount(); objIdx++) {
- LocalAbstractObject object = data.objects.getObject(objIdx);
-
- List<Integer> candidates = new ArrayList<>(); // stores the id of a candidate group
- List<Integer> overlap = new ArrayList<>();
- List<Integer> result = getSuitableObjectIdxs(object, data.objects, queryType, threshold, null);
-
- for (Integer index : result) {
- int group = data.groupIndexes.get(index);
- if (group != -1) { // if a group has already been assigned
- overlap.add(group); // store group indexes
- }
- }
- Set<Integer> distinctOverlap = new HashSet<>(overlap);
- for (Integer groupIndex : distinctOverlap) {
- // check whether each object in the group satisfies the threshold
- if (Collections.frequency(overlap, groupIndex) == data.groups.get(groupIndex).size())
- candidates.add(groupIndex);
- }
- if (candidates.isEmpty()) {
- // we need to create new group and create mapping from object to group
- Set<Integer> group = new HashSet<>();
- group.add(objIdx);
- data.groups.add(group);
- data.groupIndexes.put(objIdx, data.groups.size() - 1); // we've created a new group, hence the index is size-1
- }
- else if (candidates.size() == 1) {
- data.groups.get(candidates.get(0)).add(objIdx);
- data.groupIndexes.put(objIdx, candidates.get(0));
- }
- else {
- switch (overlapType) {
- case JOIN_ANY: {
- data.groups.get(candidates.get(0)).add(objIdx);
- data.groupIndexes.put(objIdx, candidates.get(0));
- break;
- }
- case FORM_NEW_GROUP: {
- Set<Integer> group = new HashSet<>();
- group.add(objIdx);
- data.groups.add(group);
- data.groupIndexes.put(objIdx, data.groups.size() - 1);
- break;
- }
- case ELIMINATE: {
- //todo .. toto je fakt bleh BitSet + ignoruj "zcernane prvky" ?
- }
- }
- }
- }
- return data;
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAll_N_Shuffle.java b/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAll_N_Shuffle.java
deleted file mode 100644
index 14c4eef..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAll_N_Shuffle.java
+++ /dev/null
@@ -1,74 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.groupBy;
-
-import cz.muni.fi.disa.similarityoperators.cover.AbstractRepresentation;
-import cz.muni.fi.disa.similarityoperators.utils.ObjectGroups;
-import cz.muni.fi.disa.similarityoperators.utils.OverlapType;
-import cz.muni.fi.disa.similarityoperators.utils.QueryType;
-import messif.objects.LocalAbstractObject;
-
-import java.util.*;
-
-public class SGBAll_N_Shuffle extends SGBQuery {
-
- public ObjectGroups data; // object storing objects and computed clusters
-
- public SGBAll_N_Shuffle(AbstractRepresentation.PrecomputedDistances objects) {
- data = new ObjectGroups(objects);
- }
-
- public ObjectGroups executeQuery(QueryType queryType, double threshold, OverlapType overlapType) {
-
- for (int objIdx = 0; objIdx < data.objects.getObjectCount(); objIdx++) {
- // instead of retrieving object on position i, selects object on position in shuffleArray[i]
- LocalAbstractObject object = data.objects.getObject(data.shuffleArray[objIdx]);
-
- List<Integer> candidates = new ArrayList<>(); // stores the id of a candidate group
- List<Integer> overlap = new ArrayList<>();
- List<Integer> result = getSuitableObjectIdxs(object, data.objects, queryType, threshold, data.shuffleArray);
-
- for (Integer index : result) {
- int group = data.groupIndexes.get(index);
- if (group != -1) { // if a group has already been assigned
- overlap.add(group); // store group indexes
- }
- }
- Set<Integer> distinctOverlap = new HashSet<>(overlap);
- for (Integer groupIndex : distinctOverlap) {
- // check whether each object in the group satisfies the threshold
- if (Collections.frequency(overlap, groupIndex) == data.groups.get(groupIndex).size())
- candidates.add(groupIndex);
- }
- if (candidates.isEmpty()) {
- // we need to create new group and create mapping from object to group
- Set<Integer> group = new HashSet<>();
- group.add(objIdx); // priadaj objekt
- data.groups.add(group); //pridaj group
- data.groupIndexes.put(objIdx, data.groups.size() - 1); // we've created a new group, hence the index is size-1
- }
- else if (candidates.size() == 1) {
- data.groups.get(candidates.get(0)).add(objIdx);
- data.groupIndexes.put(objIdx, candidates.get(0));
- }
- else {
- switch (overlapType) {
- case JOIN_ANY: {
- data.groups.get(candidates.get(0)).add(objIdx);
- data.groupIndexes.put(objIdx, candidates.get(0));
- break;
- }
- case FORM_NEW_GROUP: {
- Set<Integer> group = new HashSet<>();
- group.add(objIdx);
- data.groups.add(group);
- data.groupIndexes.put(objIdx, data.groups.size() - 1);
- break;
- }
- case ELIMINATE: {
- //todo .. toto je fakt bleh BitSet + ignoruj "zcernane prvky" ?
- }
- }
- }
- }
- return data;
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAny.java b/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAny.java
deleted file mode 100644
index 3c7ed06..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAny.java
+++ /dev/null
@@ -1,46 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.groupBy;
-
-import cz.muni.fi.disa.similarityoperators.cover.AbstractRepresentation.PrecomputedDistances;
-import cz.muni.fi.disa.similarityoperators.utils.QueryType;
-import cz.muni.fi.disa.similarityoperators.utils.UnionFind;
-import messif.objects.LocalAbstractObject;
-
-import java.util.List;
-
-public class SGBAny extends SGBQuery{
-
- //private QueryType queryType; // specifies the queryType
- //private UnionFind objects;
- //private double threshold; // k for the NN query or distance for the range query
-
- public List<List<LocalAbstractObject>> executeQuery(PrecomputedDistances objects,
- QueryType queryType, double threshold) {
- UnionFind objectsUnionFind = new UnionFind(objects); // creates sets, each containing a single object
-
- for (LocalAbstractObject queryObject : objects.getObjects()) {
- List<LocalAbstractObject> result = getSuitableObjects(queryObject, objectsUnionFind, queryType, threshold);
- for (LocalAbstractObject object : result) {
- // there cannot exist two groups with intersecting objects..therefore merge both "groups"
- objectsUnionFind.union(queryObject, object);
- }
- }
-
- //objectsUnionFind.printParent();
-
- return objectsUnionFind.getFinalGroups();
- }
-
- private List<LocalAbstractObject> getSuitableObjects(LocalAbstractObject queryObject, UnionFind objects, QueryType queryType, double threshold) {
- return getSuitableObjects(queryObject, objects.getPrecomputedDistances(), queryType, threshold);
- }
-
-// // searches for objects that are threshold-NN neighbours of the object
-// private List<LocalAbstractObject> knnQuery(LocalAbstractObject object, UnionFind objects, double threshold) {
-// PrecomputedDistances data = objects.getPrecomputedDistances();
-// return knnQuery(object, data, threshold);
-// }
-//
-// private List<LocalAbstractObject> rangeQuery(LocalAbstractObject object, UnionFind objects, double threshold) {
-// return rangeQuery(object, objects.getPrecomputedDistances(), threshold);
-// }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAny_N.java b/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAny_N.java
deleted file mode 100644
index b291690..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBAny_N.java
+++ /dev/null
@@ -1,50 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.groupBy;
-
-import cz.muni.fi.disa.similarityoperators.cover.AbstractRepresentation.PrecomputedDistances;
-import cz.muni.fi.disa.similarityoperators.utils.ObjectGroups;
-import cz.muni.fi.disa.similarityoperators.utils.QueryType;
-import cz.muni.fi.disa.similarityoperators.utils.UnionFind;
-import messif.objects.LocalAbstractObject;
-
-import java.util.List;
-
-public class SGBAny_N extends SGBQuery{
-
- public ObjectGroups data; // object storing objects and computed clusters
-
-
- public SGBAny_N(PrecomputedDistances objects) {
- data = new ObjectGroups(objects);
- }
- //private QueryType queryType; // specifies the queryType
- //private UnionFind objects;
- //private double threshold; // k for the NN query or distance for the range query
-
- public ObjectGroups executeQuery(QueryType queryType, double threshold) {
- UnionFind objectsUnionFind = new UnionFind(data.objects); // creates sets, each containing a single object
-
- for (LocalAbstractObject queryObject : data.objects.getObjects()) {
- List<LocalAbstractObject> result = getSuitableObjects(queryObject, objectsUnionFind, queryType, threshold);
- for (LocalAbstractObject object : result) {
- // there cannot exist two groups with intersecting objects..therefore merge both "groups"
- objectsUnionFind.union(queryObject, object);
- }
- }
-
- return objectsUnionFind.createGrouping();
- }
-
- private List<LocalAbstractObject> getSuitableObjects(LocalAbstractObject queryObject, UnionFind objects, QueryType queryType, double threshold) {
- return getSuitableObjects(queryObject, objects.getPrecomputedDistances(), queryType, threshold);
- }
-
-// // searches for objects that are threshold-NN neighbours of the object
-// private List<LocalAbstractObject> knnQuery(LocalAbstractObject object, UnionFind objects, double threshold) {
-// PrecomputedDistances data = objects.getPrecomputedDistances();
-// return knnQuery(object, data, threshold);
-// }
-//
-// private List<LocalAbstractObject> rangeQuery(LocalAbstractObject object, UnionFind objects, double threshold) {
-// return rangeQuery(object, objects.getPrecomputedDistances(), threshold);
-// }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBQuery.java b/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBQuery.java
deleted file mode 100644
index d5b3a4e..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBQuery.java
+++ /dev/null
@@ -1,192 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.groupBy;
-
-import cz.muni.fi.disa.similarityoperators.cover.AbstractRepresentation;
-import cz.muni.fi.disa.similarityoperators.cover.AbstractRepresentation.PrecomputedDistances;
-import cz.muni.fi.disa.similarityoperators.utils.ObjectGroups;
-import cz.muni.fi.disa.similarityoperators.utils.QueryType;
-import messif.objects.LocalAbstractObject;
-
-import java.io.BufferedWriter;
-import java.io.IOException;
-import java.util.*;
-
-public abstract class SGBQuery {
-
- protected List<LocalAbstractObject> getSuitableObjects(LocalAbstractObject queryObject, PrecomputedDistances objects,
- QueryType queryType, double threshold) {
- if (queryType == QueryType.RANGE) {
- return rangeQuery(queryObject, objects, threshold);
- }
- return knnQuery(queryObject, objects, threshold);
- }
-
- /**
- * get indexes of suitable objects with respect to the shuffle array
- */
- protected List<Integer> getSuitableObjectIdxs(LocalAbstractObject queryObject, PrecomputedDistances objects,
- QueryType queryType, double threshold, Integer[] shuffleArray) {
- if (queryType == QueryType.RANGE) {
- if (shuffleArray == null){
- return rangeQueryIndexes(queryObject, objects, threshold);
- }
- return rangeQueryIndexes(queryObject, objects, threshold, shuffleArray);
- }
- return knnQueryIndexes(queryObject, objects, threshold);
- }
-
-
- protected List<LocalAbstractObject> knnQuery(LocalAbstractObject object, PrecomputedDistances data, double threshold) {
- int objectIdx = data.getIndex(object);
- float[] dists = data.getDistances(objectIdx);
-
- List<Integer> res = knnQuery(objectIdx, dists, threshold);
- List<LocalAbstractObject> knnObjects = new ArrayList<>();
- for (int i = 0; i < res.size(); i++) {
- knnObjects.add(data.getObject(res.get(i)));
- }
- return knnObjects;
- }
-
- protected List<Integer> knnQuery(int objectIdx, float[] dists, double threshold) {
-
- AbstractRepresentation.Ranked[] ranked = AbstractRepresentation.Ranked.create(dists);
- Arrays.sort(ranked);
- List<Integer> res = new ArrayList<>((int)threshold);
- int i = 0;
- while (res.size() < Math.min((int) threshold, ranked.length)) {
- res.add(ranked[i].index);
- i++;
- }
- return res;
- }
-
- /** returns indexes of knn-neighbours*/
- protected List<Integer> knnQueryIndexes(LocalAbstractObject object, PrecomputedDistances data,
- double threshold) {
- int objectIdx = data.getIndex(object);
- float[] dists = data.getDistances(objectIdx);
-
- return knnQuery(objectIdx, dists, threshold);
- }
-
- protected List<LocalAbstractObject> rangeQuery(LocalAbstractObject object, PrecomputedDistances data, double threshold) {
- List<LocalAbstractObject> result = new ArrayList<>();
- for (LocalAbstractObject other : data.getObjects()) {
- if (data.getDistance(object, other) <= threshold) {
- result.add(other);
- }
- }
- return result;
- }
-
- protected List<Integer> rangeQueryIndexes(LocalAbstractObject object, PrecomputedDistances data,
- double threshold) {
- List<Integer> result = new ArrayList<>();
- int objectIndex = data.getIndex(object);
- for(int i = 0; i < data.getObjectCount(); i++) {
- if (data.getDistance(objectIndex, i) <= threshold ) {
- result.add(i);
- }
- }
- return result;
- }
-
- protected List<Integer> rangeQueryIndexes(LocalAbstractObject object, PrecomputedDistances data,
- double threshold, Integer[] shuffleArray) {
- List<Integer> result = new ArrayList<>();
- int objectIndex = data.getIndex(object); // toto je uz spravny objekt
- for(int i = 0; i < data.getObjectCount(); i++) {
- if (data.getDistance(objectIndex, shuffleArray[i]) <= threshold ) { // hladam susedov podla ich indexov v shuffle array
- result.add(i); // vrati indexy objektov odpovedajuce indexom v shuffle array
- }
- }
- return result;
- }
-
- protected List<Integer> getGroupIdxsWithObject(LocalAbstractObject object,
- List<Set<LocalAbstractObject>> groups) {
- List<Integer> result = new ArrayList<>();
- for (int i = 0; i < groups.size(); i++) {
- if (groups.get(i).contains(object))
- result.add(i);
- }
- return result;
- }
-
- protected List<List<LocalAbstractObject>> castInnerSetToList(List<Set<LocalAbstractObject>> groups) {
- List<List<LocalAbstractObject>> result = new ArrayList<>();
- for (Set<LocalAbstractObject> group : groups) {
- result.add(new ArrayList<>(group));
- }
- return result;
- }
-
-// public String toShortString(List<List<LocalAbstractObject>> groups) {
-// StringBuilder sb = new StringBuilder();
-// sb.append("GROUPS count: ").append(groups.size()).append("\n");
-// for (int i = 0; i < groups.size(); i++) {
-// sb.append("Group: ").append(i).append(" size: ").append(groups.get(i).size()).append("\n");
-// }
-// return sb.toString();
-// }
-
- public String toShortStringN(List<Set<Integer>> groups) {
- StringBuilder sb = new StringBuilder();
- sb.append("GROUPS count: ").append(groups.size()).append("\n");
- for (int i = 0; i < groups.size(); i++) {
- sb.append("Group: ").append(i).append(" size: ").append(groups.get(i).size()).append("\n");
- }
- return sb.toString();
- }
-
-// public void write(List<List<LocalAbstractObject>> groups) throws IOException {
-// System.out.println("GROUPS count: " + groups.size());
-// StringBuilder sb = new StringBuilder();
-// for (int i = 0; i < groups.size(); i++) {
-// System.out.println("Group " + i + " size: " + groups.get(i).size() + "\n" + "objects: ");
-// for (LocalAbstractObject object : groups.get(i)) {
-// object.write(System.out);
-// }
-// }
-// }
-
- public void write(ObjectGroups objectGroups, BufferedWriter writer) throws IOException {
- String objectNameConst = "#objectKey messif.objects.keys.AbstractObjectKey ";
- System.out.println("GROUPS count: " + objectGroups.groups.size());
- StringBuilder sb = new StringBuilder();
- for (int i = 0; i < objectGroups.groups.size(); i++) {
- //System.out.println("Group " + i + " size: " + groups.get(i).size() + "\n" + "objects: ");
- for (int j = 0; j < objectGroups.groups.get(i).size(); j++) {
- LocalAbstractObject object = objectGroups.objects.getObject(objectGroups.shuffleArray[j]);
- sb.append(objectNameConst).append(object.getLocatorURI()).append("\n").append(i).append("\n"); // object identifier and group separated by \n
- }
- }
- writer.write(sb.toString());
- //System.out.println(sb.toString());
- }
-
- public void write(HashMap<Integer, List<Integer>> mapping, SoftSGBStrategies clustering, BufferedWriter writer) throws IOException {
- String objectNameConst = "#objectKey messif.objects.keys.AbstractObjectKey ";
- System.out.println("GROUPS count: " + clustering.mainGrouping.groups.size());
- StringBuilder sb = new StringBuilder();
- for (Map.Entry<Integer, List<Integer>> objClusterPair : mapping.entrySet()) {
- LocalAbstractObject object = clustering.mainGrouping.objects.getObject
- (clustering.mainGrouping.shuffleArray[objClusterPair.getKey()]);
- sb.append(objectNameConst).append(object.getLocatorURI()).append("\n"); // object identifier and group separated by \n
- boolean flag = false;
- for (Integer clusterId : objClusterPair.getValue()){
- if (flag) {
- sb.append(",").append(clusterId);
- }
- else{
- sb.append(clusterId);
- flag = true;
- }
- }
- sb.append("\n");
- }
- writer.write(sb.toString());
- //System.out.println(sb.toString());
- }
-
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBVote.java b/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBVote.java
deleted file mode 100644
index aaf8536..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBVote.java
+++ /dev/null
@@ -1,51 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.groupBy;
-
-import cz.muni.fi.disa.similarityoperators.cover.AbstractRepresentation;
-import cz.muni.fi.disa.similarityoperators.utils.QueryType;
-import messif.objects.LocalAbstractObject;
-
-import java.util.ArrayList;
-import java.util.HashSet;
-import java.util.List;
-import java.util.Set;
-
-public class SGBVote extends SGBQuery {
-
- public List<List<LocalAbstractObject>> executeQuery(AbstractRepresentation.PrecomputedDistances objects,
- QueryType queryType, double threshold) {
- List<Set<LocalAbstractObject>> groups = new ArrayList<>();
-
- for (LocalAbstractObject queryObject : objects.getObjects()) {
- List<LocalAbstractObject> result = getSuitableObjects(queryObject, objects, queryType, threshold);
- float[] weights = new float[groups.size()];
- for (LocalAbstractObject o : result) {
- List<Integer> groupsIndexes = getGroupIdxsWithObject(o, groups);
- if (!groupsIndexes.isEmpty()) {
- if (objects.getDistance(queryObject, o) == 0f) {
- weights[groupsIndexes.get(0)] = Float.MAX_VALUE;
- } else {
- weights[groupsIndexes.get(0)] += (1 / objects.getDistance(queryObject, o));
- }
- }
- }
- int maxIdx = -1;
- float maxWeight = 0f;
- for (int i = 0; i < weights.length; i++) {
- if (weights[i] != 0 && weights[i] > maxWeight) {
- maxIdx = i;
- maxWeight = weights[i];
- }
- }
-
- if (maxIdx == -1){
- Set<LocalAbstractObject> newGroup = new HashSet<>();
- newGroup.add(queryObject);
- groups.add(newGroup);
- }
- else {
- groups.get(maxIdx).add(queryObject);
- }
- }
- return castInnerSetToList(groups);
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBVote_N.java b/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBVote_N.java
deleted file mode 100644
index 833c359..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBVote_N.java
+++ /dev/null
@@ -1,61 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.groupBy;
-
-import cz.muni.fi.disa.similarityoperators.cover.AbstractRepresentation;
-import cz.muni.fi.disa.similarityoperators.utils.ObjectGroups;
-import cz.muni.fi.disa.similarityoperators.utils.QueryType;
-import messif.objects.LocalAbstractObject;
-
-import java.util.HashSet;
-import java.util.List;
-import java.util.Set;
-
-public class SGBVote_N extends SGBQuery {
-
- public ObjectGroups data; // object storing objects and computed clusters
-
- public SGBVote_N(AbstractRepresentation.PrecomputedDistances objects) {
- data = new ObjectGroups(objects);
- }
-
- public ObjectGroups executeQuery(QueryType queryType, double threshold) {
-
- for (int objIdx = 0; objIdx < data.objects.getObjectCount(); objIdx++) {
- LocalAbstractObject queryObject = data.objects.getObject(objIdx);
-
- List<Integer> result = getSuitableObjectIdxs(queryObject, data.objects, queryType, threshold, null);
- float[] weights = new float[data.groups.size()];
-
- for (Integer otherIdx : result) {
- int group = data.groupIndexes.get(otherIdx);
-
- if (group != -1) { // object has already a group assigned
- if (data.objects.getDistance(objIdx, otherIdx) == 0f) {
- weights[group] = Float.MAX_VALUE;
- } else {
- weights[group] += (1 / data.objects.getDistance(objIdx, otherIdx));
- }
- }
- }
- int maxIdx = -1;
- float maxWeight = 0f;
- for (int i = 0; i < weights.length; i++) {
- if (weights[i] != 0 && weights[i] > maxWeight) {
- maxIdx = i;
- maxWeight = weights[i];
- }
- }
-
- if (maxIdx == -1){
- Set<Integer> newGroup = new HashSet<>();
- newGroup.add(objIdx);
- data.groups.add(newGroup);
- data.groupIndexes.put(objIdx, data.groups.size() - 1);
- }
- else {
- data.groups.get(maxIdx).add(objIdx);
- data.groupIndexes.put(objIdx, maxIdx);
- }
- }
- return data;
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBVote_N_Shuffle.java b/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBVote_N_Shuffle.java
deleted file mode 100644
index ae2a809..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGBVote_N_Shuffle.java
+++ /dev/null
@@ -1,63 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.groupBy;
-
-import cz.muni.fi.disa.similarityoperators.cover.AbstractRepresentation;
-import cz.muni.fi.disa.similarityoperators.utils.ObjectGroups;
-import cz.muni.fi.disa.similarityoperators.utils.QueryType;
-import messif.objects.LocalAbstractObject;
-
-import java.util.HashSet;
-import java.util.List;
-import java.util.Set;
-
-public class SGBVote_N_Shuffle extends SGBQuery {
-
- public ObjectGroups data; // object storing objects and computed clusters
-
- public SGBVote_N_Shuffle(AbstractRepresentation.PrecomputedDistances objects) {
- data = new ObjectGroups(objects);
- }
-
- public ObjectGroups executeQuery(QueryType queryType, double threshold) {
-
- for (int objIdx = 0; objIdx < data.objects.getObjectCount(); objIdx++) {
-
- LocalAbstractObject queryObject = data.objects.getObject(data.shuffleArray[objIdx]);
-
- List<Integer> result = getSuitableObjectIdxs(queryObject, data.objects, queryType, threshold, data.shuffleArray);
- float[] weights = new float[data.groups.size()];
-
- for (Integer otherIdx : result) {
- int group = data.groupIndexes.get(otherIdx); // odpoveda group risluchajucej indexu v shuffle array
-
- if (group != -1) { // object has already a group assigned
- // najdi prvky podla indexov v shuffle array
- if (data.objects.getDistance(data.shuffleArray[objIdx], data.shuffleArray[otherIdx]) == 0f) {
- weights[group] = Float.MAX_VALUE;
- } else {
- weights[group] += (1 / data.objects.getDistance(data.shuffleArray[objIdx], data.shuffleArray[otherIdx]));
- }
- }
- }
- int maxIdx = -1;
- float maxWeight = 0f;
- for (int i = 0; i < weights.length; i++) {
- if (weights[i] != 0 && weights[i] > maxWeight) {
- maxIdx = i;
- maxWeight = weights[i];
- }
- }
-
- if (maxIdx == -1){
- Set<Integer> newGroup = new HashSet<>();
- newGroup.add(objIdx);
- data.groups.add(newGroup);
- data.groupIndexes.put(objIdx, data.groups.size() - 1);
- }
- else {
- data.groups.get(maxIdx).add(objIdx);
- data.groupIndexes.put(objIdx, maxIdx);
- }
- }
- return data;
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGB_All_Any.java b/src/cz/muni/fi/disa/similarityoperators/groupBy/SGB_All_Any.java
deleted file mode 100644
index 1ad94cd..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/groupBy/SGB_All_Any.java
+++ /dev/null
@@ -1,6 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.groupBy;
-
-public class SGB_All_Any {
-
-
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/groupBy/SoftSGBStrategies.java b/src/cz/muni/fi/disa/similarityoperators/groupBy/SoftSGBStrategies.java
deleted file mode 100644
index bc5190d..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/groupBy/SoftSGBStrategies.java
+++ /dev/null
@@ -1,259 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.groupBy;
-import cz.muni.fi.disa.similarityoperators.cover.*;
-import cz.muni.fi.disa.similarityoperators.utils.ObjectGroups;
-import messif.objects.LocalAbstractObject;
-
-import java.util.*;
-
-public class SoftSGBStrategies extends SGBQuery{
-
- public ObjectGroups mainGrouping;
-
- private int MAX_NN;
-
- public SoftSGBStrategies(ObjectGroups grouping) {
- this.mainGrouping = grouping;
- }
-
- // porovnavam vyhradne vzdialenosti medoidov voci sebe..
- // dostatocne blizke clustre spojim tak, ze vsetkym objektom v clustri nastavim ako dalsie cluster_ids prave clusterIds tychto medoidov
- public HashMap<Integer, List<Integer>> knnClusteringStrategy(double thresholdDist, int maxNNs) {
- this.MAX_NN = maxNNs;
- // vracaj indexy nie objekty
- int[] medoids = computeMedoidsIdxs();
- HashMap<Integer, List<Integer>> softClustering = new HashMap<>();
-
- for(int i = 0; i < medoids.length; i++){
- // search for other object within the arrayList for objects within thresholdDist
- int mClusterId = this.mainGrouping.groupIndexes.get(i);
- List<Integer> clusterIds = new ArrayList<>();
- clusterIds.add(mClusterId);
- List<Integer> kNNClusterIds = getClusterIdsOfNNsWithinThreshold(i, thresholdDist, medoids);
- clusterIds.addAll(kNNClusterIds);
- for (Integer clusterObjectIdx : mainGrouping.groups.get(mClusterId)){
- softClustering.put(clusterObjectIdx, clusterIds);
- }
- }
- return softClustering;
- }
-
- // pre kazdy objekt hladam blizke medoidy..
- // dostatocne blizke clustre spojim tak, ze vsetkym objektom v clustri nastavim ako dalsie cluster_ids prave clusterIds tychto medoidov
- public HashMap<Integer, List<Integer>> knnMedoidsClusteringStrategy(double thresholdDist, int maxNNs) {
- // vracaj indexy nie objekty
- this.MAX_NN = maxNNs;
- int[] medoidIdxs = computeMedoidsIdxs();
- HashMap<Integer, List<Integer>> softClustering = new HashMap<>();
-
- for(int i = 0; i < this.mainGrouping.objects.getObjectCount(); i++){
- // search for other object within the arrayList for objects within thresholdDist
- // je mi jedno ci to spravim v poradi shuffle array alebo objektov v precomp dists
- int mClusterId = this.mainGrouping.groupIndexes.get(i);
- List<Integer> clusterIds = new ArrayList<>();
- clusterIds.add(mClusterId);
- List<Integer> kNNclusterIds = getClusterIdsOfNNsWithinThreshold(i, thresholdDist, medoidIdxs);
- clusterIds.addAll(kNNclusterIds);
- softClustering.put(i, clusterIds);
- }
- return softClustering;
- }
-
-
- public HashMap<Integer, List<Integer>> hullClusteringStrategy(double thresholdDist, int maxNNs) {
- this.MAX_NN = maxNNs;
- HashMap<Integer, List<Integer>> softClustering = new HashMap<>();
- int[] hullObjectsIdxs = getHullIdxs();
-
- // pre kazy objekt vyber najblizsich kNN hullObjektov blizsich ako threshold
- for (int i = 0; i < this.mainGrouping.objects.getObjectCount(); i++) {
- int mClusterId = this.mainGrouping.groupIndexes.get(i);
- List<Integer> clusterIds = new ArrayList<>();
- clusterIds.add(mClusterId);
- List<Integer> kNNclusterIds = getClusterIdsOfNNsWithinThreshold(i, thresholdDist,hullObjectsIdxs );
- clusterIds.addAll(kNNclusterIds);
- softClustering.put(i, clusterIds);
- }
- return softClustering;
- }
-
- private int[] getHullIdxs() {
- List<Integer> hullIndexes = new ArrayList<>();
- for (Set<Integer> cluster : this.mainGrouping.groups) {
- Map<LocalAbstractObject, Integer> clusterObjects = getClusterObjects(cluster);
- HullRepresentation hull = new HullOptimizedRepresentationV3(new ArrayList<>(clusterObjects.keySet())).build();
- List<LocalAbstractObject> hullObjects = hull.getHull();
-
- for (LocalAbstractObject o : hullObjects) { // preiteruje vsetky objekty a najde indexy odpovedajuce hullObjektom
- hullIndexes.add(clusterObjects.get(o));
- }
- }
- return hullIndexes.stream().mapToInt(i->i).toArray();
- }
-
- private Map<LocalAbstractObject, Integer> getClusterObjects(Set<Integer> clusterIdxs) {
- Map<LocalAbstractObject, Integer> result = new HashMap<>();
- for (Integer idx : clusterIdxs) {
- result.put(this.mainGrouping.objects.getObject(this.mainGrouping.shuffleArray[idx]), idx);
- }
- return result;
- }
-
-
- public HashMap<Integer, List<Integer>> categoriesClusteringStrategies(double thresholdDist, int maxNNs) {
- this.MAX_NN = maxNNs;
- HashMap<Integer, List<Integer>> softClustering = new HashMap<>();
-
- for(int i = 0; i < this.mainGrouping.objects.getObjectCount(); i++){
- // search for other objects within the thresholdDist
- List<Integer> objectsIdxWithinThreshold = rangeQueryWithinIndexesExcludingMyClusterObject(i, thresholdDist);
-
- // mam blizke clustre vratane seba, teda vytvor "soft"
- List<Integer> selectedKnnClusterIds = selectKnnWithDifferentCategories(i, objectsIdxWithinThreshold);
-
- // najdi si clusterIds pre pozice knnMedoids
- // pre vsetky objekty v m-clustri nastav soft :P
- List<Integer> clusterIds = new ArrayList<>();
- int oClusteId = this.mainGrouping.groupIndexes.get(i);
- clusterIds.add(oClusteId);
- clusterIds.addAll(selectedKnnClusterIds);
-
- softClustering.put(i, clusterIds);
- }
- return softClustering;
- }
-
- private List<Integer> selectKnnWithDifferentCategories(int objectIdx, List<Integer> candidateIdxs) {
- // zaprve usporiadaj podla vzdialeosti.. nasledne parsni nazov aa vyber Set kategorii pokym nie je velky ako k
- LocalAbstractObject object = this.mainGrouping.objects.getObject(this.mainGrouping.shuffleArray[objectIdx]);
- String objectCategory = getObjectsCategory(object);
-
- float[] dists = selectDistsToObjects(objectIdx, candidateIdxs);
- AbstractRepresentation.Ranked[] ranked = AbstractRepresentation.Ranked.create(dists);
- Arrays.sort(ranked);
- Set<Integer> selectedClusterIds = new HashSet<>();
- selectedClusterIds.add(this.mainGrouping.groupIndexes.get(objectIdx));
- int i = 0;
- while (selectedClusterIds.size() != MAX_NN && i < ranked.length){
- int candidateIdx = candidateIdxs.get(ranked[i].index);
- LocalAbstractObject candidate = this.mainGrouping.objects.getObject(this.mainGrouping.shuffleArray[candidateIdx]);
- String candidateCategory = getObjectsCategory(candidate);
- if (candidateCategory.equals(objectCategory) &&
- !this.mainGrouping.groupIndexes.get(objectIdx).equals(this.mainGrouping.groupIndexes.get(candidateIdx))) {
- selectedClusterIds.add(this.mainGrouping.groupIndexes.get(candidateIdx));
- }
- i++;
- }
- selectedClusterIds.remove(this.mainGrouping.groupIndexes.get(objectIdx));
- return new ArrayList<>(selectedClusterIds);
- }
-
- private String getObjectsCategory(LocalAbstractObject object) {
- String objectIdentifier = object.getLocatorURI();
- String[] splits = objectIdentifier.split("_");
- return splits[1];
- }
-
- private List<Integer> selectKnnIndexesWithinArray(int objectIdx, List<Integer> selectedIndexes) {
-
- return getKnnClusterIds(objectIdx, selectedIndexes); // vrati indexy vramci selectedIdxs
- }
-
- private float[] selectDistsToObjects(int objectIdx, List<Integer> selectedIndexes) {
- float[] dists = new float[selectedIndexes.size()];
- for (int i = 0; i < selectedIndexes.size(); i++){
- dists[i] = this.mainGrouping.objects.getDistance(
- this.mainGrouping.shuffleArray[objectIdx],
- this.mainGrouping.shuffleArray[selectedIndexes.get(i)]);
- }
- return dists;
- }
-
- // vrati indexy medoidov with respect to the shuffle array
- private int[] computeMedoidsIdxs() {
- int[] medoidsIdxs = new int[mainGrouping.groups.size()]; // wrt shuffle array
- for (int i = 0; i < mainGrouping.groups.size(); i++) {
- Set<Integer> cluster = mainGrouping.groups.get(i); // ziskaj cluster
- // sum dists for each object of the clusters and pick the min
- int medoidIdx = getMedoidIdx(cluster);
- medoidsIdxs[i] = medoidIdx;
- }
-
- return medoidsIdxs;
- }
-
- private int getMedoidIdx(Set<Integer> cluster) {
- double minSum = Double.MAX_VALUE;
- int medoidIdx = -1; // some object will always be assigned, since minDist cannot be
- for (int candidateIdx : cluster) {
- double objectSumDists = 0;
- for (int otherIdx : cluster) {
- objectSumDists += Math.pow(mainGrouping.objects.getDistance(mainGrouping.shuffleArray[candidateIdx],
- mainGrouping.shuffleArray[otherIdx]),2); //
- }
- if (objectSumDists < minSum) {
- minSum = objectSumDists;
- medoidIdx = candidateIdx;
- }
- }
- return medoidIdx;
- }
-
- private List<Integer> getClusterIdsOfNNsWithinThreshold(int objIdx, double thresholdDist, int[] medoidsIdxs) {
-
- List<Integer> nearMedoids = rangeQueryWithinIndexes(objIdx, thresholdDist, medoidsIdxs);
- // mam blizke clustre vratane seba, teda vytvor "soft"
- // dostanem clusterIds pr kNNs
- List<Integer> knnClusterIDs = selectKnnIndexesWithinArray(objIdx, nearMedoids);
- // vratane seba
-
- return knnClusterIDs;
- }
-
- private List<Integer> rangeQueryWithinIndexes(int objectIdx, double threshold, int[] indexes) {
- List<Integer> result = new ArrayList<>();
- //int objectIndex = data.getIndex(object);
- for(int i = 0; i < indexes.length; i++) {
- if (this.mainGrouping.objects.getDistance(
- this.mainGrouping.shuffleArray[objectIdx],
- this.mainGrouping.shuffleArray[indexes[i]]) <= threshold) {
- result.add(indexes[i]);
- }
- }
- return result;
- }
-
- private List<Integer> rangeQueryWithinIndexesExcludingMyClusterObject(int objectIdx, double threshold) {
- List<Integer> result = new ArrayList<>();
- //int objectIndex = data.getIndex(object);
- for(int i = 0; i < this.mainGrouping.objects.getObjectCount(); i++) {
- if (this.mainGrouping.objects.getDistance(
- this.mainGrouping.shuffleArray[objectIdx],
- this.mainGrouping.shuffleArray[i]) <= threshold) {
- if (!this.mainGrouping.groupIndexes.get(objectIdx).equals(this.mainGrouping.groupIndexes.get(i))){
- result.add(i);
- }
- }
- }
- return result;
- }
-
- private List<Integer> getKnnClusterIds(int objectIdx, List<Integer> candidateIdxs ){
-
- float[] dists = selectDistsToObjects(objectIdx, candidateIdxs);
- AbstractRepresentation.Ranked[] ranked = AbstractRepresentation.Ranked.create(dists);
- Arrays.sort(ranked);
- Set<Integer> selectedClusterIds = new HashSet<>();
- selectedClusterIds.add(this.mainGrouping.groupIndexes.get(objectIdx));
- int i = 0;
- while (selectedClusterIds.size() != MAX_NN && i < ranked.length){
- int candidateIdx = candidateIdxs.get(ranked[i].index);
- if (!this.mainGrouping.groupIndexes.get(objectIdx).equals(this.mainGrouping.groupIndexes.get(candidateIdx))) {
- selectedClusterIds.add(this.mainGrouping.groupIndexes.get(candidateIdx));
- }
- i++;
- }
- selectedClusterIds.remove(this.mainGrouping.groupIndexes.get(objectIdx));
- return new ArrayList<>(selectedClusterIds);
- }
-
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/operations/Containment.java b/src/cz/muni/fi/disa/similarityoperators/operations/Containment.java
deleted file mode 100644
index 2c5bba5..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/operations/Containment.java
+++ /dev/null
@@ -1,22 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.operations;
-
-import cz.muni.fi.disa.similarityoperators.cover.HullRepresentation;
-import messif.objects.LocalAbstractObject;
-
-public class Containment {
-
- public static boolean isContainedInHull(HullRepresentation hull,
- LocalAbstractObject object) {
- //int[] hullIdxs = hull.getHullIdxs();
- //return hull.isExternalPointCoveredByHull(object, hullIdxs, hullIdxs.length);
- return hull.isExternalCovered(object);
- }
-
- public static boolean isContainedInHull(HullRepresentation hull,
- LocalAbstractObject object,
- int k) {
- //int[] hullIdxs = hull.getHullIdxs();
- //return hull.isExternalPointCoveredByHull(object, hullIdxs, k);
- return hull.isExternalCovered(object, k);
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/operations/Intersection.java b/src/cz/muni/fi/disa/similarityoperators/operations/Intersection.java
deleted file mode 100644
index 5cb2514..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/operations/Intersection.java
+++ /dev/null
@@ -1,157 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.operations;
-
-import cz.muni.fi.disa.similarityoperators.cover.*;
-import messif.objects.LocalAbstractObject;
-
-import java.io.FileNotFoundException;
-import java.io.FileOutputStream;
-import java.io.IOException;
-import java.io.OutputStream;
-import java.util.ArrayList;
-import java.util.List;
-
-public class Intersection {
-
- // probably will not be used
- public static HullRepresentation computeIntersectionV1(
- HullRepresentation firstHull, HullRepresentation secondHull) {
-
- List<LocalAbstractObject> firstHullData = firstHull.getObjects();
- List<LocalAbstractObject> secondHullData = secondHull.getObjects();
-
- return computeIntersectionShared(firstHullData, secondHullData,
- firstHull, secondHull,
- "", 0, 0);
- }
-
- // computes intersection on HullPoints
- public static HullRepresentation computeIntersection(HullMHPRepresentation firstHull,
- HullMHPRepresentation secondHull) {
- List<LocalAbstractObject> firstHullData = firstHull.getHull();
- List<LocalAbstractObject> secondHullData = secondHull.getHull();
- return computeIntersectionShared(firstHullData, secondHullData,
- firstHull, secondHull, "", 3, 0);
- }
-
- public static HullRepresentation computeIntersection(HullOptimizedRepresentation firstHull,
- HullOptimizedRepresentation secondHull) {
- List<LocalAbstractObject> firstHullData = firstHull.getHull();
- List<LocalAbstractObject> secondHullData = secondHull.getHull();
- return computeIntersectionShared(firstHullData, secondHullData,
- firstHull, secondHull, "", 0, 0);
- }
-
- public static HullRepresentation computeIntersection(HullKNHPRepresentation firstHull,
- HullKNHPRepresentation secondHull) {
- List<LocalAbstractObject> firstHullData = firstHull.getHull();
- List<LocalAbstractObject> secondHullData = secondHull.getHull();
- return computeIntersectionShared(firstHullData, secondHullData,
- firstHull, secondHull, "", firstHull.getMinHullPoints(), firstHull.getkNearestHullPoints());
- }
-
- public static HullRepresentation computeIntersection(HullOptimizedRepresentationV2 firstHull,
- HullOptimizedRepresentationV2 secondHull) {
- List<LocalAbstractObject> firstHullData = firstHull.getHull();
- List<LocalAbstractObject> secondHullData = secondHull.getHull();
- return computeIntersectionShared(firstHullData, secondHullData,
- firstHull, secondHull, "V2", 0, 0);
- }
-
- public static HullRepresentation computeIntersection(HullRepresentation firstHull,
- HullRepresentation secondHull) {
- List<LocalAbstractObject> firstHullData = firstHull.getHull();
- List<LocalAbstractObject> secondHullData = secondHull.getHull();
- return computeIntersectionShared(firstHullData, secondHullData,
- firstHull, secondHull, "", 0, 0);
- }
-
- private static HullRepresentation computeIntersectionShared(List<LocalAbstractObject> firstHullData,
- List<LocalAbstractObject> secondHullData,
- HullRepresentation firstHull,
- HullRepresentation secondHull,
- String version,
- int minHullPoints,
- int kNHullPoints) {
- List<LocalAbstractObject> coveredData = new ArrayList<>();
-
- //each point is covered by points in firstHull, we are checking for coverage using secondHull
- getHullCoveredObjects(firstHullData, secondHull, coveredData, kNHullPoints);
- getHullCoveredObjects(secondHullData, firstHull, coveredData, kNHullPoints);
-
- // based on parameters create correct version on Hull representation class
- if (version.equals("V2"))
- return new HullOptimizedRepresentationV2(coveredData).build();
- if (kNHullPoints != 0)
- return new HullKNHPRepresentation(coveredData, 3, kNHullPoints).build();
- if (minHullPoints != 0)
- return new HullMHPRepresentation(coveredData, minHullPoints).build();
- return new HullRepresentation(coveredData).build();
- }
-
- // method checking whether intersection between two hulls exists
- public static boolean doesIntersectionExist(HullRepresentation firstHull,
- HullRepresentation secondHull,
- int k){
- List<LocalAbstractObject> coveredList = new ArrayList<LocalAbstractObject>();
-
- List<LocalAbstractObject> firstHullData = firstHull.getHull();
- List<LocalAbstractObject> secondHullData = secondHull.getHull();
-
- getHullCoveredObjects(firstHullData, secondHull, coveredList, k);
- getHullCoveredObjects(secondHullData, firstHull, coveredList, k);
- return !coveredList.isEmpty();
- }
-
- public static void getHullCoveredObjects(List<LocalAbstractObject> hullSharedData,
- AbstractHullRepresentation hull,
- List<LocalAbstractObject> coveredData,
- int kNHullPoints){
- int[] hullIdxs = hull.getHullIdxs();
-
- for(LocalAbstractObject hullPoint : hullSharedData){
- int hullPoints = kNHullPoints == 0 ? hullIdxs.length : kNHullPoints;
- if (hull.isExternalCovered(hullPoint, hullPoints, hullIdxs)) {
- // insert point only if it is not already present in covered data ..
- if (!isPointInList(hullPoint, coveredData))
- coveredData.add(hullPoint); //point from first hull covered by point in second hull
- }
- }
- }
-
- private static boolean isPointInList(LocalAbstractObject obj, List<LocalAbstractObject> covered){
- for (LocalAbstractObject cov : covered) {
- if (cov.dataEquals(obj)){
- return true;
- }
- }
- return false;
- }
-
- public static void printIntersectionPoints(HullRepresentation first, HullRepresentation second) {
- System.out.println("Points in intersection");
- try {
- OutputStream out = new FileOutputStream("./intersection" + ".txt", false);
- for( LocalAbstractObject obj : first.getObjects()) {
- if (second.isExternalCovered(obj, second.getRepresentativesCount(), second.getHullIdxs())){
- try {
- obj.write(out);
- } catch (IOException e) {
- e.printStackTrace();
- }
- }
- }
- for( LocalAbstractObject obj : second.getObjects()) {
- if (first.isExternalCovered(obj, first.getRepresentativesCount(), first.getHullIdxs())){
- try {
- obj.write(out);
- } catch (IOException e) {
- e.printStackTrace();
- }
- }
- }
- } catch (FileNotFoundException e) {
- e.printStackTrace();
- }
-
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/operations/Subset.java b/src/cz/muni/fi/disa/similarityoperators/operations/Subset.java
deleted file mode 100644
index be92707..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/operations/Subset.java
+++ /dev/null
@@ -1,23 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.operations;
-
-import cz.muni.fi.disa.similarityoperators.cover.HullRepresentation;
-import messif.objects.LocalAbstractObject;
-
-import java.util.ArrayList;
-import java.util.List;
-
-public class Subset {
-
- /**
- * checks whether first hull is subset of the other
- */
- public static boolean isSubset(HullRepresentation firstHull,
- HullRepresentation secondHull,
- int k) {
- List<LocalAbstractObject> coveredList = new ArrayList<>();
- List<LocalAbstractObject> firstHullData = firstHull.getHull();
-
- Intersection.getHullCoveredObjects(firstHullData, secondHull, coveredList, k);
- return coveredList.size() == firstHullData.size();
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/operations/Union.java b/src/cz/muni/fi/disa/similarityoperators/operations/Union.java
deleted file mode 100644
index 94481df..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/operations/Union.java
+++ /dev/null
@@ -1,152 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.operations;
-
-import cz.muni.fi.disa.similarityoperators.cover.*;
-import cz.muni.fi.disa.similarityoperators.cover.AbstractRepresentation.PrecomputedDistances;
-import messif.objects.LocalAbstractObject;
-
-import java.io.IOException;
-import java.util.ArrayList;
-import java.util.List;
-
-// "static" class in java, final = no extensions possible
-public final class Union {
-
- //private default constructor, instances cannot be created
- private Union() {}
-
- //Computes union on hull points KNHP
- public static HullRepresentation ComputeUnionOnHullPoints(
- HullKNHPRepresentation firstHull, HullKNHPRepresentation secondHull) {
- List<LocalAbstractObject> hullPoints = new ArrayList<>();
-
- hullPoints.addAll(firstHull.getHull());
- hullPoints.addAll(secondHull.getHull());
-
- return new HullKNHPRepresentation(
- hullPoints,
- firstHull.getMinHullPoints() ,
- firstHull.getkNearestHullPoints())
- .build();
- }
-
- // union on Hulls - min hp
- public static HullRepresentation ComputeUnionOnHullPoints(
- HullMHPRepresentation firstHull, HullMHPRepresentation secondHull) {
- List<LocalAbstractObject> hullPoints = new ArrayList<>();
-
- hullPoints.addAll(firstHull.getHull());
- hullPoints.addAll(secondHull.getHull());
-
- return new HullMHPRepresentation(
- hullPoints,
- firstHull.getMinHullPoints())
- .build();
- }
-
- public static HullRepresentation ComputeUnionOnHullPoints(
- HullOptimizedRepresentationV2 firstHull, HullOptimizedRepresentationV2 secondHull) {
- List<LocalAbstractObject> hullPoints = new ArrayList<>();
-
- hullPoints.addAll(firstHull.getHull());
- hullPoints.addAll(secondHull.getHull());
-
- return new HullOptimizedRepresentationV2(hullPoints).build();
- }
-
- public static HullRepresentation ComputeUnionOnHullPoints(
- HullOptimizedRepresentation firstHull, HullOptimizedRepresentation secondHull) {
- List<LocalAbstractObject> hullPoints = new ArrayList<>();
-
- hullPoints.addAll(firstHull.getHull());
- hullPoints.addAll(secondHull.getHull());
-
- return new HullOptimizedRepresentation(hullPoints).build();
- }
-
- public static HullRepresentation ComputeUnionOnHullPoints(HullRepresentation firstHull,
- HullRepresentation secondHull) {
- List<LocalAbstractObject> hullPoints = new ArrayList<>();
-
- hullPoints.addAll(firstHull.getHull());
- hullPoints.addAll(secondHull.getHull());
-
- return new HullRepresentation(hullPoints).build();
- }
-
-
- //Computes union of hulls - used when having only single picture //
- public static HullRepresentation ComputeUnionOnData(PrecomputedDistances shared, String version) {
- if (version.equals("V2"))
- return new HullOptimizedRepresentationV2(shared).build();
- return new HullOptimizedRepresentation(shared).build();
- }
-
- public static HullRepresentation ComputeUnionOnData(PrecomputedDistances shared,
- int minHullPoints) {
- return new HullMHPRepresentation(shared, minHullPoints).build();
- }
-
- public static HullRepresentation ComputeUnionOnData(PrecomputedDistances shared,
- int minHullPoints, int kNHullPoints) {
- return new HullKNHPRepresentation(shared, minHullPoints, kNHullPoints).build();
- }
-
- public static HullRepresentation ComputeUnionOnData(PrecomputedDistances shared) {
- return new HullRepresentation(shared).build();
- }
-
- //hullIndexes found by hand in shared
- public static boolean CheckCoveredByHull(PrecomputedDistances shared, HullRepresentation union,
- int kNHullPoints){
-
- int kNN = kNHullPoints == 0 || kNHullPoints > union.getRepresentativesCount() ? union.getRepresentativesCount() : kNHullPoints;
-
- for (int i = 0; i < shared.getObjectCount(); i++){
-// if (!union.isExternalPointCoveredByHull(shared.getObject(i), union.getHullIdxs(), kNN)) {
- if (!union.isExternalCovered(shared.getObject(i), kNN)) {
- try {
- shared.getObject(i).write(System.out);
- } catch (IOException e) {
- e.printStackTrace();
- }
- //System.out.println("not covered " + i);
- //return false;
- }
- }
- return true;
- }
-
-// public static boolean CheckCoveredByHull(PrecomputedDistances shared, HullRepresentation union,
-// int kNHullPoints){
-// HullRepresentation hull = new HullRepresentation(shared);
-// int[] idxs = mapHullPointsToSharedData(shared, union); //maps each point in hull into shared
-// System.out.println(Arrays.toString(idxs));
-//
-// int kNN = kNHullPoints == 0 || kNHullPoints < union.getRepresentativesCount() ? union.getRepresentativesCount() : kNHullPoints;
-//
-//
-// for (int i = 0; i < shared.getObjectCount(); i++){
-// if (!hull.isCoveredByHull(i, kNN, idxs)) {
-// System.out.println("not covered " + i);
-// //return false;
-// }
-// }
-// return true;
-// }
-//
-//
-// public static int[] mapHullPointsToSharedData(PrecomputedDistances points,
-// HullRepresentation hull){
-// int[] hullIndexes = new int[hull.getHull().size()];
-// int nextHullIndex = 0;
-// for (LocalAbstractObject hullPoint : hull.getHull()) {
-// for (int i = 0; i < points.getObjectCount(); i++) {
-// if (hullPoint.dataEquals(points.getObject(i))) {
-// hullIndexes[nextHullIndex] = i;
-// nextHullIndex ++;
-// }
-// }
-// }
-// return hullIndexes;
-// }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/utils/ImageNet.java b/src/cz/muni/fi/disa/similarityoperators/utils/ImageNet.java
deleted file mode 100644
index ce8b2e0..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/utils/ImageNet.java
+++ /dev/null
@@ -1,89 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.utils;
-
-import java.io.*;
-import java.util.HashMap;
-import java.util.Map;
-import java.util.logging.Level;
-import java.util.logging.Logger;
-import java.util.zip.GZIPInputStream;
-
-/**
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class ImageNet {
- private static final String MAP_SERIALIZED_FILE = "id-to-url.bin";
- private static Map<String,String> idToUrl = null;
-
- public static String urlForId(String objectLocator) {
- if (idToUrl == null)
- readIdToUrlMap();
-
- String url = idToUrl.get(objectLocator);
- if (url == null) {
- objectLocator = objectLocator.replaceFirst("\\..*$", "");
- url = idToUrl.get(objectLocator);
- }
- if (url == null) {
- objectLocator = objectLocator.replaceFirst("^.*/", "");
- url = idToUrl.get(objectLocator);
- }
- return url;
- }
-
-
- public static void readIdToUrlMap() {
- File f = new File(MAP_SERIALIZED_FILE);
- if (f.canRead()) {
- try {
- ObjectInputStream in = new ObjectInputStream(new FileInputStream(f));
- idToUrl = (Map)in.readObject();
- in.close();
- return;
- } catch (IOException | ClassNotFoundException ex) {
- Logger.getLogger(ImageNet.class.getName()).log(Level.SEVERE, null, ex);
- }
- }
-
- readIdToUrlMap("//andromeda/share/datasets/imagenet/imagenet_fall11_urls.txt.gz");
-
- if (!f.canRead()) {
- try {
- ObjectOutputStream out = new ObjectOutputStream(new FileOutputStream(f));
- out.writeObject(idToUrl);
- out.close();
- } catch (IOException ex) {
- Logger.getLogger(ImageNet.class.getName()).log(Level.SEVERE, null, ex);
- }
- }
- }
-
- public static void readIdToUrlMap(String filePath) {
- idToUrl = new HashMap<>();
- try {
- GZIPInputStream gzipIn = new GZIPInputStream(new FileInputStream(filePath));
- BufferedReader in = new BufferedReader(new InputStreamReader(gzipIn));
-
- while (true) {
- String line = in.readLine();
- if (line == null || line.isEmpty())
- break;
- String[] parts = line.split("[ \\t]+", 2);
- if (parts.length == 2)
- idToUrl.put(parts[0], parts[1]);
- if ((idToUrl.size() % 100000) == 0)
- System.out.println("Map size: " + idToUrl.size());
- }
- in.close();
- } catch (FileNotFoundException ex) {
- Logger.getLogger(ImageNet.class.getName()).log(Level.SEVERE, null, ex);
- } catch (IOException ex) {
- Logger.getLogger(ImageNet.class.getName()).log(Level.SEVERE, null, ex);
- }
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/utils/MergingType.java b/src/cz/muni/fi/disa/similarityoperators/utils/MergingType.java
deleted file mode 100644
index 9c84c5a..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/utils/MergingType.java
+++ /dev/null
@@ -1,7 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.utils;
-
-public enum MergingType {
- NONE,
- BASIC,
- SUM_BASED
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/utils/ObjectGroups.java b/src/cz/muni/fi/disa/similarityoperators/utils/ObjectGroups.java
deleted file mode 100644
index f7cef61..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/utils/ObjectGroups.java
+++ /dev/null
@@ -1,33 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.utils;
-
-import cz.muni.fi.disa.similarityoperators.cover.AbstractRepresentation;
-
-import java.util.*;
-
-public class ObjectGroups {
-
- public AbstractRepresentation.PrecomputedDistances objects;
-
- /** For shuffling purposes of the precomputed distances matrix */
- public Integer[] shuffleArray;
-
- /** Mapping from ids in precomputed distances to group (i.e. cluster) the object belongs to */
- public HashMap<Integer, Integer> groupIndexes; //
- /** clusters */
- public List<Set<Integer>> groups ;
-
- public ObjectGroups(AbstractRepresentation.PrecomputedDistances objects) {
- this.objects = objects;
- groupIndexes = new HashMap<>();
- groups = new ArrayList<>();
- this.shuffleArray = new Integer[objects.getObjectCount()];
- // initialization of grouping ..
- for (int i = 0; i < objects.getObjectCount(); i++) {
- groupIndexes.put(i, -1);
- shuffleArray[i] = i;
- }
- List<Integer> arrayList = Arrays.asList(shuffleArray);
- Collections.shuffle(arrayList);
- arrayList.toArray(shuffleArray);
- }
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/utils/OverlapType.java b/src/cz/muni/fi/disa/similarityoperators/utils/OverlapType.java
deleted file mode 100644
index f4d97bf..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/utils/OverlapType.java
+++ /dev/null
@@ -1,7 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.utils;
-
-public enum OverlapType {
- JOIN_ANY,
- FORM_NEW_GROUP,
- ELIMINATE;
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/utils/QueryType.java b/src/cz/muni/fi/disa/similarityoperators/utils/QueryType.java
deleted file mode 100644
index 93dc7fe..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/utils/QueryType.java
+++ /dev/null
@@ -1,6 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.utils;
-
-public enum QueryType {
- kNN,
- RANGE
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/utils/UnionFind.java b/src/cz/muni/fi/disa/similarityoperators/utils/UnionFind.java
deleted file mode 100644
index 6dbdba0..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/utils/UnionFind.java
+++ /dev/null
@@ -1,126 +0,0 @@
-package cz.muni.fi.disa.similarityoperators.utils;
-
-import cz.muni.fi.disa.similarityoperators.cover.AbstractRepresentation.PrecomputedDistances;
-import messif.objects.LocalAbstractObject;
-
-import java.util.*;
-
-// Union-find data structure with link by rank and path compression
-// therefore combination of m makeSet, find and union operations over the set of n elements takes O(m * @alpha(n)) time.
-public class UnionFind {
-
- //private Map<LocalAbstractObject, Integer> mapping; // object and its index
- private PrecomputedDistances objects; // the data structure itself stores indexes for each object
- private byte[] rank;
- private int[] parent;
- private int numberOfGroups;
-
-
- // initializes empty union-find data structure
- // each element is in its own set
- public UnionFind(PrecomputedDistances objects) {
- //this.mapping = new HashMap<>();
- this.rank = new byte[objects.getObjectCount()];
- this.parent = new int[objects.getObjectCount()];
- for (int i = 0; i < objects.getObjectCount(); i++){
- makeSet(objects.getObject(i), i);
- }
- this.objects = objects;
- }
-
- public void makeSet(LocalAbstractObject x, int i) {
- rank[i] = 0;
- parent[i] = i;
- //mapping.put(x, i);
- }
-
- // implementation of find with path compression
- public LocalAbstractObject find(LocalAbstractObject x) { // finds the root of the tree elem belongs to
- //int index = mapping.get(x);
- int index = objects.getIndex(x);
- if (index != parent[index]) {
- LocalAbstractObject grandParent = find(objects.getObject(parent[index])); // find the root
- parent[index] = objects.getIndex(grandParent); // get grandparent index
- }
- return objects.getObject(parent[index]);
- }
-
- public int getNumberOfGroups(){
- return numberOfGroups;
- }
-
- public List<LocalAbstractObject> getObjects() {
- return this.objects.getObjects();
- }
-
- public PrecomputedDistances getPrecomputedDistances() { return this.objects; }
-
- //
- public void union(LocalAbstractObject x, LocalAbstractObject y) {
- LocalAbstractObject rootX = find(x);
- LocalAbstractObject rootY = find(y);
- int rootXIdx = objects.getIndex(rootX);
- int rootYIdx = objects.getIndex(rootY);
-
- if (rootXIdx == rootYIdx) return; // already in the same group
-
- // link by rank
- else if (rank[rootXIdx] < rank[rootYIdx])
- parent[rootXIdx] = rootYIdx;
- else if (rank[rootXIdx] > rank[rootYIdx])
- parent[rootYIdx] = rootXIdx;
- else { // the rank is the same, therefore increase rank of the new root
- parent[rootYIdx] = rootXIdx; // can be switched by parent[rootXIdx] = rootYIdx
- rank[rootXIdx]++;
- }
- numberOfGroups--;
- }
-
- public ObjectGroups createGrouping() {
- ObjectGroups data = new ObjectGroups(objects);
-
- //Map<Integer, List<LocalAbstractObject>> groups = new HashMap<>();
-
- for (int i = 0; i < this.parent.length; i++) { // loop cez vsetky prvky
- int parentIdx = parent[i];
- Integer clusterIdx = data.groupIndexes.get(parentIdx);
- LocalAbstractObject queryObject = data.objects.getObject(i);
- if (clusterIdx == -1) { // no group assigned yet for this parent
- Set<Integer> newGroup = new HashSet<>();
- newGroup.add(i);
- data.groups.add(newGroup);
- data.groupIndexes.put(i, data.groups.size() - 1);
- data.groupIndexes.put(parentIdx, data.groups.size()-1);
- }
- else {
- data.groups.get(clusterIdx).add(i);
- data.groupIndexes.put(i, clusterIdx);
- }
- }
-
- return data;
- }
-
- public List<List<LocalAbstractObject>> getFinalGroups() {
- Map<Integer, List<LocalAbstractObject>> groups = new HashMap<>();
- for (int i = 0; i < this.parent.length; i++) {
- int parentIdx = parent[i];
- List<LocalAbstractObject> cluster = groups.get(parentIdx);
- if (cluster == null) { // no entry for this parent
- cluster = new ArrayList<>();
- }
- cluster.add(objects.getObject(i)); // add objects index
- groups.put(parentIdx, cluster); // value is replaced
- }
-
- return new ArrayList<>(groups.values());
-
- }
-
- public void printParent() {
- for (int i =0; i < parent.length; i++){
- System.out.println(i + " parent - " + parent[i]);
- }
- }
-
-}
diff --git a/src/cz/muni/fi/disa/similarityoperators/utils/UnsupervisedBinaryClassification.java b/src/cz/muni/fi/disa/similarityoperators/utils/UnsupervisedBinaryClassification.java
deleted file mode 100644
index 76c5a98..0000000
--- a/src/cz/muni/fi/disa/similarityoperators/utils/UnsupervisedBinaryClassification.java
+++ /dev/null
@@ -1,289 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package cz.muni.fi.disa.similarityoperators.utils;
-
-import java.util.ArrayList;
-import java.util.Iterator;
-import java.util.List;
-
-/**
- * Manages binary classification of pairs of objects in totals in TP, FP, TN and FN categories.
- *
- * The pairs are judged to fall into a category by threshold distances (similarity, dissimilarity).
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class UnsupervisedBinaryClassification implements Iterable<UnsupervisedBinaryClassification.BinaryClassification> {
- List<BinaryClassification> items;
-
- /** Create a list of binary classification for updating all by passing one distance at once.
- * @param thresholds a list of elements of even length, even (from 0) positions are similarity thresholds and odd positions are dissimilarity thresholds
- */
- public UnsupervisedBinaryClassification(float... thresholds) {
- items = new ArrayList<>(thresholds.length / 2);
- for (int i = 0; i < thresholds.length; i += 2)
- items.add(new BinaryClassification(thresholds[i], thresholds[i+1]));
- }
-
- /** Soft-assingment binary classificaation.
- * Binary-classify the passed distance between a pair of objects that are assigned to cluster1 and to soft clusters (2), respectively.
- * @param pairDist distance between objects
- * @param cluster1 cluster ID of the first object
- * @param clusters2 array of cluster IDs of the second object
- */
- public void classify(float pairDist, List<Integer> cluster1, List<Integer> clusters2) {
- for (BinaryClassification bc : items) {
- bc.classify(pairDist, cluster1, clusters2);
- }
- }
-
- /** Soft-assingment binary classificaation.
- * Binary-classify the passed distance between a pair of objects that are assigned to cluster1 and to soft clusters (2), respectively.
- * @param pairDist distance between objects
- * @param cluster1 cluster ID of the first object
- * @param clusters2 array of cluster IDs of the second object
- */
- public void classifyAnd(float pairDist, List<Integer> cluster1, List<Integer> clusters2) {
- for (BinaryClassification bc : items) {
- bc.classifyAnd(pairDist, cluster1, clusters2);
- }
- }
-
- /** Binary-classify the passed distance between a pair of objects that are assigned to cluster1 and cluster2 clusters IDs, respectively.
- * @param pairDist distance between objects
- * @param cluster1 cluster ID of the first object
- * @param cluster2 cluster ID of the second object
- */
- public void classify(float pairDist, int cluster1, int cluster2) {
- for (BinaryClassification bc : items) {
- bc.classify(pairDist, cluster1, cluster2);
- }
- }
-
- /** Binary-classify the passed distance between a pair of objects that match to the passed degree (from 0 (non-matching) to 1 (perfectly matching)).
- * @param pairDist distance between objects
- * @param matchConfidence confidence of match for the pair of objects
- */
- public void classify(float pairDist, float matchConfidence) {
- for (BinaryClassification bc : items) {
- bc.classify(pairDist, matchConfidence);
- }
- }
-
- @Override
- public String toString() {
- StringBuilder sb = new StringBuilder();
- for (BinaryClassification bc : items) {
- sb.append(bc).append("\n");
- }
- return sb.toString();
- }
-
- @Override
- public Iterator<BinaryClassification> iterator() {
- return items.iterator();
- }
-
- public static class BinaryClassification {
- /** Threshold on distance to mark a pair (distance) as similar (close) */
- private final float similar;
- /** Threshold on distance to mark a pair (distance) as dissimilar (far apart) */
- private final float dissimilar;
-
- /** True positive count */
- private long tp;
- /** False positive count */
- private long fp;
- /** True negative count */
- private long tn;
- /** False negative count */
- private long fn;
-
- /** Number of pairs in the grey zone -- non-distiquishable by threholds */
- private long greyZone = 0;
-
- /** Grey area counts -- they are weighted */
- private double tpGrey = 0;
- private double fpGrey = 0;
- private double tnGrey = 0;
- private double fnGrey = 0;
-
- public BinaryClassification(float similarThreshold, float dissimilarThreshold) {
- this.similar = similarThreshold;
- this.dissimilar = dissimilarThreshold;
- }
-
- public void classify(float pairDist, int cluster1, int cluster2) {
- classify(pairDist, ((cluster1 == cluster2) ? 1f : 0f));
-// if (pairDist <= similar) {
-// if (cluster1 == cluster2)
-// tp++;
-// else
-// fn++;
-// } else if (pairDist > dissimilar) {
-// if (cluster1 != cluster2)
-// tn++;
-// else
-// fp++;
-// } else { // ignore the distance between the thresholds -- we cannot distiguish them without the real GT.
-// greyZone++;
-// double w = weightDistance(pairDist);
-// if (cluster1 == cluster2) {
-// // Same cluster, so contribute to TP and FP
-// tpGrey += (1.0d - w);
-// fpGrey += w;
-// } else {
-// // Diffrent clusters, so contribute to TN and FN
-// tnGrey += w;
-// fnGrey += (1.0d - w);
-// }
-// assert (greyZone < 10000 || (greyZone-1) < (long)(tpGrey + fpGrey + tnGrey + fnGrey)) : "Grey zone classification inconsistency: greyZone=" + greyZone + ", tpGrey=" + tpGrey + ", fpGrey=" + fpGrey + ", tnGrey=" + tnGrey + ", fnGrey=" + fnGrey;
-// }
- }
-
- /** Soft-assignement classification. ... v podstate odpoveda OR-klasifikaci, lebo sledujem obe strany a staci ze najdem jeden match
- * It might not be symmetric to compare the objects, so we have to check both the directions here!!!
- */
- public void classify(float pairDist, List<Integer> clusters1, List<Integer> clusters2) {
- // Test for a match from o1->o2
- for (Integer c2 : clusters2) {
- if (clusters1.get(0).equals(c2)) {
- classify(pairDist, 1f);
- return;
- }
- }
- // Test for a match from o2->o1
- for (Integer c1 : clusters1) {
- if (clusters2.get(0).equals(c1)) {
- classify(pairDist, 1f);
- return;
- }
- }
- classify(pairDist, 0f);
- }
-
- /** Soft-assignement classification. ... v podstate odpoveda AND-klasifikaci, lebo sledujem obe strany a staci ze najdem jeden match
- * It might not be symmetric to compare the objects, so we have to check both the directions here!!!
- */
- public void classifyAnd(float pairDist, List<Integer> clusters1, List<Integer> clusters2) {
- // Test for a match from o1->o2
- boolean doubleMatch = false;
- for (Integer c2 : clusters2) {
- if (clusters1.get(0).equals(c2)) {
- doubleMatch = true;
- break;
- }
- }
- if (doubleMatch) {
- // Test for a match from o2->o1
- for (Integer c1 : clusters1) {
- if (clusters2.get(0).equals(c1)) {
- classify(pairDist, 1f);
- return;
- }
- }
- }
-
- classify(pairDist, 0f);
- }
-
- /** Binary-classify the passed distance between a pair of objects that match to the passed degree (from 0 (non-matching) to 1 (perfectly matching)).
- * @param pairDist distance between objects
- * @param matchConfidence confidence of match for the pair of objects
- */
- public void classify(float pairDist, float matchConfidence) {
- if (pairDist <= similar) {
- if (matchConfidence > 0)
- tp++;
- else
- fn++;
- } else if (pairDist > dissimilar) {
- if (matchConfidence == 0)
- tn++;
- else
- fp++;
- } else { // ignore the distance between the thresholds -- we cannot distiguish them without the real GT.
- greyZone++;
- double w = weightDistance(pairDist);
- if (matchConfidence > 0) {
- // Same cluster, so contribute to TP and FP
- tpGrey += (1.0d - w);
- fpGrey += w;
- } else {
- // Diffrent clusters, so contribute to TN and FN
- tnGrey += w;
- fnGrey += (1.0d - w);
- }
- assert (greyZone < 10000 || (greyZone-1) < (long)(tpGrey + fpGrey + tnGrey + fnGrey)) : "Grey zone classification inconsistency: greyZone=" + greyZone + ", tpGrey=" + tpGrey + ", fpGrey=" + fpGrey + ", tnGrey=" + tnGrey + ", fnGrey=" + fnGrey;
- }
- }
-
- private double weightDistance(double dist) {
- return (dist - similar) / (dissimilar - similar);
- }
-
- @Override
- public String toString() {
- return "BinaryClassification{similar=" + similar + ", dissimilar=" + dissimilar + ": tp=" + tp + ", fp=" + fp + ", tn=" + tn + ", fn=" + fn
- + ", greyZone=" + greyZone + ", tpGrey=" + tpGrey + ", fpGrey=" + fpGrey + ", tnGrey=" + tnGrey + ", fnGrey=" + fnGrey + '}';
- }
-
- public long getCorrectDecisions() {
- return tp + tn;
- }
-
- public double getCorrectDecisionsWithGreyZone() {
- return tp + tn + tpGrey + tnGrey;
- }
-
- public long getIncorrectDecisions() {
- return fp + fn;
- }
-
- public double getIncorrectDecisionsWithGreyZone() {
- return fp + fn + fpGrey + fnGrey;
- }
-
- public double getRandIndex() {
- return (double)getCorrectDecisions() / (double)(getCorrectDecisions() + getIncorrectDecisions());
- }
-
- public double getRandIndexWithGreyZone() {
- return getCorrectDecisionsWithGreyZone() / (getCorrectDecisionsWithGreyZone() + getIncorrectDecisionsWithGreyZone());
- }
-
- public float getAdjustedRandIndex(BinaryClassification rndCf) {
- return (float)(getCorrectDecisions() - rndCf.getCorrectDecisions())
- / (float)(getCorrectDecisions() + getIncorrectDecisions() - rndCf.getCorrectDecisions());
- }
-
- public double getAdjustedRandIndexWithGreyZone(BinaryClassification rndCf) {
- return (getCorrectDecisionsWithGreyZone() - rndCf.getCorrectDecisionsWithGreyZone())
- / (getCorrectDecisionsWithGreyZone() + getIncorrectDecisionsWithGreyZone() - rndCf.getCorrectDecisionsWithGreyZone());
- }
-
- public double getPrecision() {
- return (double)tp / (double)(tp + fn);
- }
-
- public double getRecall() {
- return (double)tp / (double)(tp + fp);
- }
-
- public double getFscore() {
- return 2d * getPrecision() * getRecall() / (getPrecision() + getRecall());
- }
-
- public double getFBscore(double beta) {
- return (1d + beta * beta) * getPrecision() * getRecall() / ((beta * beta * getPrecision()) + getRecall());
- }
-
- /** Fowlkes–Mallows index */
- public double getGmeasure() {
- return Math.sqrt(getPrecision() * getRecall());
- }
- }
-}
diff --git a/src/messif/objects/keys/RadiusObjectKey.java b/src/messif/objects/keys/RadiusObjectKey.java
deleted file mode 100644
index e945c7f..0000000
--- a/src/messif/objects/keys/RadiusObjectKey.java
+++ /dev/null
@@ -1,186 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package messif.objects.keys;
-
-import messif.objects.nio.BinaryInput;
-import messif.objects.nio.BinaryOutput;
-import messif.objects.nio.BinarySerializator;
-
-import java.io.IOException;
-import java.io.OutputStream;
-
-/**
- * This class adds to the standard object key {@link AbstractObjectKey} radius of the ball region that this object represents.
- * So a float value is stored.
- * Radius is stored before the URI (locator)! E.g. #objectKey messif.objects.keys.RadiusObjectKey [6.234] all_souls_000000.jpg
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class RadiusObjectKey extends AbstractObjectKey {
- /** Class serial id for serialization. */
- private static final long serialVersionUID = 1L;
-
- //************ Attributes ************//
-
- /** Radius of a ball */
- private final float radius;
-
-
- //************ Constructor ************//
-
- /**
- * Creates a new instance for a string value.
- * @param value the string representaion of the key
- */
- public RadiusObjectKey(String value) {
- // Radius is stored before the URI (locator)! E.g. #objectKey messif.objects.keys.RadiusObjectKey [6.234] all_souls_000000.jpg
- super(AbstractObjectKey.getKeyStringPart(value, " ", 1));
- // Dimensions:
- try {
- String d = AbstractObjectKey.getKeyStringPart(value, " ", 0).substring(value.indexOf('[') + 1, value.indexOf(']'));
- this.radius = Float.parseFloat(d);
- } catch (NumberFormatException ex) {
- throw new IllegalArgumentException("Incorrect radius object key format! Expected: \"[radius] URI\" Passed: " + value, ex);
- }
- }
-
- /**
- * Creates a new instance.
- * @param locatorURI the URI locator
- * @param radius ball radius value
- */
- public RadiusObjectKey(String locatorURI, float radius) {
- super(locatorURI);
- this.radius = radius;
- }
-
- //************ Attribute access methods ************//
-
- /**
- * @return value of the radius
- */
- public float getRadius() {
- return radius;
- }
-
- //****************** Serialization ******************//
-
- /**
- * Store this key's data to a text stream.
- * This method should have the opposite deserialization in constructor.
- * Note that this method should <em>not</em> write a line separator (\n).
- *
- * @param stream the stream to store this object to
- * @throws IOException if there was an error while writing to stream
- */
- @Override
- protected void writeData(OutputStream stream) throws IOException {
- stream.write(formatDimensionsToString().getBytes());
- super.writeData(stream);
- }
-
- private String formatDimensionsToString() {
- return "[" + radius + "]";
- }
-
- //************ Comparator and equality methods ************//
-
- /**
- * Compare the keys according to their locators.
- * @param o the key to compare this key with
- * @return a negative integer, zero, or a positive integer if this object
- * is less than, equal to, or greater than the specified object
- */
- @Override
- public int compareTo(AbstractObjectKey o) {
- if (o == null || !(o.getClass().equals(RadiusObjectKey.class)))
- return 3;
- int cmp = super.compareTo(o);
- if (cmp != 0)
- return cmp;
-
- RadiusObjectKey obj = (RadiusObjectKey)o;
- return Float.compare(this.radius, obj.radius);
- }
-
- /**
- * Return the hashCode of the locator URI or 0, if it is null.
- * @return the hashCode of the locator URI or 0, if it is null
- */
- @Override
- public int hashCode() {
- return super.hashCode() * 47 + Float.hashCode(radius);
- }
-
- /**
- * Returns whether this key is equal to the <code>obj</code>.
- * It is only and only if the <code>obj</code> is descendant of
- * {@link AbstractObjectKey} and has an equal locator URI.
- *
- * @param obj the object to compare this object to
- * @return <tt>true</tt> if this object is the same as the <code>obj</code> argument; <tt>false</tt> otherwise
- */
- @Override
- public boolean equals(Object obj) {
- if (!super.equals(obj))
- return false;
- if (!(obj.getClass().equals(RadiusObjectKey.class)))
- return false;
-
- RadiusObjectKey o = (RadiusObjectKey)obj;
- return (o.radius == this.radius);
- }
-
-
- //************ String representation ************//
-
- /**
- * Returns the URI string.
- * @return the URI string
- */
- @Override
- public String toString() {
- return formatDimensionsToString() + super.toString();
- }
-
-
- //************ BinarySerializable interface ************//
-
- /**
- * Creates a new instance of AbstractObjectKey loaded from binary input.
- *
- * @param input the input to read the AbstractObjectKey from
- * @param serializator the serializator used to write objects
- * @throws IOException if there was an I/O error reading from the input
- */
- protected RadiusObjectKey(BinaryInput input, BinarySerializator serializator) throws IOException {
- super(input, serializator);
- this.radius = serializator.readFloat(input);
- }
-
- /**
- * Binary-serialize this object into the <code>output</code>.
- * @param output the output that this object is binary-serialized into
- * @param serializator the serializator used to write objects
- * @return the number of bytes actually written
- * @throws IOException if there was an I/O error during serialization
- */
- @Override
- public int binarySerialize(BinaryOutput output, BinarySerializator serializator) throws IOException {
- return super.binarySerialize(output, serializator) + serializator.write(output, radius);
- }
-
- /**
- * Returns the exact size of the binary-serialized version of this object in bytes.
- * @param serializator the serializator used to write objects
- * @return size of the binary-serialized version of this object
- */
- @Override
- public int getBinarySize(BinarySerializator serializator) {
- return super.getBinarySize(serializator) + serializator.getBinarySize(radius);
- }
-
-}
diff --git a/src/messif/utility/FileUtils.java b/src/messif/utility/FileUtils.java
deleted file mode 100644
index 4d8555e..0000000
--- a/src/messif/utility/FileUtils.java
+++ /dev/null
@@ -1,38 +0,0 @@
-package messif.utility;
-
-
-import messif.objects.LocalAbstractObject;
-import messif.objects.util.StreamGenericAbstractObjectIterator;
-
-import java.io.IOException;
-import java.util.ArrayList;
-import java.util.List;
-
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-
-/**
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class FileUtils {
- public static <E extends LocalAbstractObject> StreamGenericAbstractObjectIterator<E> openDB(Class<E> clazz, String file) {
- try {
- return new StreamGenericAbstractObjectIterator(clazz, file);
- } catch (IllegalArgumentException | IOException ex) {
- throw new RuntimeException(ex);
- }
- }
-
- public static <E extends LocalAbstractObject> List<E> readObjects(Class<E> clazz, String file) {
- StreamGenericAbstractObjectIterator<E> iter = openDB(clazz, file);
- List<E> objs = new ArrayList<>();
- while (iter.hasNext())
- objs.add(iter.next());
- return objs;
- }
-
-}
diff --git a/src/messif/utility/HullRepresentationAsLocalAbstractObject.java b/src/messif/utility/HullRepresentationAsLocalAbstractObject.java
deleted file mode 100644
index bab6fab..0000000
--- a/src/messif/utility/HullRepresentationAsLocalAbstractObject.java
+++ /dev/null
@@ -1,107 +0,0 @@
-/*
- * To change this license header, choose License Headers in Project Properties.
- * To change this template file, choose Tools | Templates
- * and open the template in the editor.
- */
-package messif.utility;
-
-import cz.muni.fi.disa.similarityoperators.cover.HullRepresentation;
-import messif.objects.LocalAbstractObject;
-
-import java.io.IOException;
-import java.io.OutputStream;
-import java.util.Iterator;
-import java.util.Objects;
-
-/**
- * Encapsulates a hull representation as a {@link LocalAbstractObject}.
- * No distance function is implemented here yet.
- *
- * @author Vlastislav Dohnal, Masaryk University, Brno, Czech Republic, dohnal@fi.muni.cz
- */
-public class HullRepresentationAsLocalAbstractObject extends LocalAbstractObject {
-
- private HullRepresentation hull;
-
- public HullRepresentationAsLocalAbstractObject(HullRepresentation h) {
- super(h.getLocatorID());
- this.hull = h;
- }
-
- public HullRepresentation getHull() {
- return hull;
- }
-
- @Override
- public int getSize() {
- return -1;
- }
-
- @Override
- public String toString() {
- return "HullRepresentationAsLocalAbstractObject{" + hull.getLocatorID() + '}';
- }
-
- @Override
- public boolean dataEquals(Object obj) {
- if (this == obj)
- return true;
- if (!getClass().isAssignableFrom(obj.getClass()))
- return false;
- HullRepresentationAsLocalAbstractObject other = (HullRepresentationAsLocalAbstractObject)obj;
- if (hull.getRepresentativesCount() != other.hull.getRepresentativesCount())
- return false;
- final Iterator<LocalAbstractObject> it1 = hull.getRepresentatives();
- final Iterator<LocalAbstractObject> it2 = other.hull.getRepresentatives();
- while (it1.hasNext() && it2.hasNext()) {
- if (!it1.next().dataEquals(it2.next()))
- return false;
- }
- return true;
- }
-
- @Override
- public int dataHashCode() {
- int hash = 7;
- final Iterator<LocalAbstractObject> it = hull.getRepresentatives();
- while (it.hasNext())
- hash = 31 * hash + it.next().dataHashCode();
- return hash;
- }
-
- @Override
- public int hashCode() {
- int hash = 5;
- hash = 71 * hash + this.hull.getLocatorID().hashCode();
- return hash;
- }
-
- @Override
- public boolean equals(Object obj) {
- if (this == obj) {
- return true;
- }
- if (obj == null) {
- return false;
- }
- if (getClass() != obj.getClass()) {
- return false;
- }
- final HullRepresentationAsLocalAbstractObject other = (HullRepresentationAsLocalAbstractObject) obj;
- if (!Objects.equals(this.hull, other.hull)) {
- return false;
- }
- return this.hull.getLocatorID().equals(other.hull.getLocatorID());
- }
-
- @Override
- protected void writeData(OutputStream stream) throws IOException {
- throw new UnsupportedOperationException("Not supported yet."); //To change body of generated methods, choose Tools | Templates.
- }
-
- @Override
- protected float getDistanceImpl(LocalAbstractObject obj, float distThreshold) {
- throw new UnsupportedOperationException("Not supported yet."); //To change body of generated methods, choose Tools | Templates.
- }
-
-}
--
GitLab