Skip to content

Commit

Permalink
add: process-D sampling by proximity
Browse files Browse the repository at this point in the history
  • Loading branch information
@PtrMan
PtrMan committed Nov 21, 2019
1 parent 0e6473f commit 621f063
Show file tree
Hide file tree
Showing 3 changed files with 82 additions and 29 deletions.
6 changes: 0 additions & 6 deletions src/main/java/ptrman/levels/retina/LineDetectorWithMultiplePoints.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -16,8 +16,6 @@
public class LineDetectorWithMultiplePoints {
public List<ProcessA.Sample> samples = new ArrayList<>(); // actual samples which are "included" in the line

public IntList integratedSampleIndices;

// variable for the line drawing in the acceleration structure
//public ArrayRealVector spatialAccelerationLineDirection; // can be null
//public double spatialAccelerationLineLength; // can be null
Expand All @@ -44,10 +42,6 @@ public LineDetectorWithMultiplePoints(double xStep) {
this.xStep = xStep;
}

public boolean doesContainSampleIndex(int index) {
return integratedSampleIndices.contains(index);
}

/** compute sigmoid like activation function as described in
* https://www.foundalis.com/res/Generalization_of_Hebbian_Learning_and_Categorization.pdf
*/
Expand Down
98 changes: 75 additions & 23 deletions src/main/java/ptrman/levels/retina/ProcessD.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -15,11 +15,13 @@
import org.apache.commons.math3.stat.regression.SimpleRegression;
import org.eclipse.collections.api.IntIterable;
import org.eclipse.collections.api.list.primitive.IntList;
import org.eclipse.collections.api.tuple.primitive.IntIntPair;
import org.eclipse.collections.impl.list.mutable.primitive.IntArrayList;
import ptrman.Datastructures.Vector2d;
import ptrman.bpsolver.HardParameters;
import ptrman.bpsolver.Parameters;
import ptrman.levels.retina.helper.ProcessConnector;
import ptrman.math.ArrayRealVectorHelper;
import ptrman.misc.Assert;

import java.util.*;
Expand Down Expand Up @@ -109,51 +111,101 @@ public void removeCandidatesBelowActivation(double threshold) {
}
}

// tries to sample a new line candidate
public void sampleNew() {
final double maxLength = Math.sqrt(squaredDistance(new double[]{imageSize.x, imageSize.y})); // max length of line

List<LineDetectorWithMultiplePoints> multiplePointsLineDetector = new ArrayList<>();

final List<ProcessA.Sample> workingSamples = inputSampleConnector.getWorkspace();

if (workingSamples.isEmpty()) {
return;
/**
* selects random samples
* @param source
* @return
*/
public List<ProcessA.Sample> selectRandomSamples(List<ProcessA.Sample> source) {
if (source.size() < 3) {
return new ArrayList<>();
}

// pick out random points
int sampleIndex = 0; // NOTE< index in endosceletonPoint / workingSamples >
IntArrayList allCandidateSampleIndices = new IntArrayList();
for (final ProcessA.Sample iterationSample : source) {
allCandidateSampleIndices.add(sampleIndex);
sampleIndex++;
}

IntList chosenCandidateSampleIndices = getRandomElements(allCandidateSampleIndices, 3, rng);
return getSamplesByIndices(source, chosenCandidateSampleIndices);
}

/**
* tries to sample a new line candidate by picking random points
*/
public void sampleNewByRandom() {
final double maxLength = Math.sqrt(squaredDistance(new double[]{imageSize.x, imageSize.y})); // max length of line

List<LineDetectorWithMultiplePoints> multiplePointsLineDetector = new ArrayList<>();

final List<ProcessA.Sample> workingSamples = inputSampleConnector.getWorkspace();

// filter valid points
List<ProcessA.Sample> filteredSamples = new ArrayList<>();
for (final ProcessA.Sample iterationSample : workingSamples) {
boolean onlyAddEndoskeletonEnable = !(onlyEndoskeleton && iterationSample.type != ProcessA.Sample.EnumType.ENDOSCELETON);
boolean isReferenced = iterationSample.refCount != 0;
if(!isReferenced && onlyAddEndoskeletonEnable ) {
allCandidateSampleIndices.add(sampleIndex);
filteredSamples.add(iterationSample);
}
sampleIndex++;
}

List<ProcessA.Sample> selectedSamples = selectRandomSamples(filteredSamples);
tryCreateMultiLineDetector(maxLength, selectedSamples);
}

IntList chosenCandidateSampleIndices = getRandomElements(allCandidateSampleIndices, 3, rng);
List<ProcessA.Sample> selectedSamples = getSamplesByIndices(workingSamples, chosenCandidateSampleIndices);
public double processDSampleByProximityProximity = 20.0; // maximal proximity of points to get considered for sampling by proximity

/**
* tries to sample a new detector by proximity of points
*/
public void sampleNewByProximity() {
final double maxLength = Math.sqrt(squaredDistance(new double[]{imageSize.x, imageSize.y})); // max length of line

final List<ProcessA.Sample> workingSamples = inputSampleConnector.getWorkspace();

tryCreateMultiLineDetector(maxLength, chosenCandidateSampleIndices, selectedSamples);
if (workingSamples.size() < 2) {
return;
}

int centerPointIdx = rng.nextInt(workingSamples.size());
IntIntPair centerPointPos = workingSamples.get(centerPointIdx).position;

// find all points in proximity
List<ProcessA.Sample> proximitySamples = new ArrayList<>();
for(ProcessA.Sample iSample : workingSamples) {
double dist = ArrayRealVectorHelper.distance(centerPointPos, iSample.position);
if (dist > processDSampleByProximityProximity) {
continue;
}

boolean onlyAddEndoskeletonEnable = !(onlyEndoskeleton && iSample.type != ProcessA.Sample.EnumType.ENDOSCELETON);
boolean isReferenced = iSample.refCount != 0;
if(!isReferenced && onlyAddEndoskeletonEnable ) {
proximitySamples.add(iSample);
}
}

// try to create line detector by selecting random points from candidates
List<ProcessA.Sample> selectedSamples = selectRandomSamples(proximitySamples);
tryCreateMultiLineDetector(maxLength, selectedSamples);
}

public double processDNumberOfPointsToActivationScale = 0.15; // how much does a point improve the scaling

private void tryCreateMultiLineDetector(
double maxLength,
IntList chosenCandidateSampleIndices, List<ProcessA.Sample> selectedSamples) {
private void tryCreateMultiLineDetector(double maxLength, List<ProcessA.Sample> selectedSamples) {
// commented check because we don't assume object id's anymore (because it was from Phaeaco for solving BP's)
//final boolean doAllSamplesHaveId = doAllSamplesHaveObjectId(selectedSamples);
//if (!doAllSamplesHaveId) {
// return;
//}

if (selectedSamples.size() == 0) {
return; // special case
if (selectedSamples.size() < 2) {
return; // line is not defined
}

// check if object ids are the same
Expand All @@ -180,14 +232,13 @@ private void tryCreateMultiLineDetector(
}
// else we are here

if(chosenCandidateSampleIndices.size() <= 2) { // the regression mse is not defined if it are only two points
if(selectedSamples.size() <= 2) { // the regression mse is not defined if it are only two points
return; // only create detector if we have at least three samples
}


// create new line detector
LineDetectorWithMultiplePoints createdLineDetector = new LineDetectorWithMultiplePoints(lineDetectorInitialXStep);
createdLineDetector.integratedSampleIndices = chosenCandidateSampleIndices;
createdLineDetector.samples = selectedSamples;

Assert.Assert(areObjectIdsTheSameOfSamples(selectedSamples), "");
Expand All @@ -198,7 +249,7 @@ private void tryCreateMultiLineDetector(

boolean addCreatedLineDetector = false;

if (createdLineDetector.integratedSampleIndices.size() == 2) {
if (createdLineDetector.samples.size() == 2) {
createdLineDetector.mse = 0.0f;

createdLineDetector.n = regressionResult.n;
Expand Down Expand Up @@ -279,7 +330,8 @@ private static List<RetinaPrimitive> splitDetectorsIntoLines(Iterable<LineDetect
* processing step
*/
public void step() {
sampleNew();
sampleNewByRandom();
sampleNewByProximity();
tryWiden();
detectorsHarden();
detectorsFadeActivation();
Expand Down
7 changes: 7 additions & 0 deletions src/main/java/ptrman/math/ArrayRealVectorHelper.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -123,4 +123,11 @@ public static double distance(IntIntPair a, ArrayRealVector b) {
double dy = b.getEntry(1) - a.getTwo();
return Math.sqrt( dx*dx+dy*dy);
}
// TODO< rename class >
/** cartesian distance */
public static double distance(IntIntPair a, IntIntPair b) {
double dx = b.getOne() - a.getOne();
double dy = b.getTwo() - a.getTwo();
return Math.sqrt( dx*dx+dy*dy);
}
}

0 comments on commit 621f063

Please sign in to comment.