Skip to content

Commit

Permalink
Merge pull request deeplearning4j#1422 from deeplearning4j/sa_arraywr…
Browse files Browse the repository at this point in the history 
…itable

Add ArrayWritable to optimize loading of images
  • Loading branch information
@saudet
saudet committed Apr 27, 2016
2 parents d5609f8 + 30c6b51 commit 6a0efff
Show file tree
Hide file tree
Showing 10 changed files with 169 additions and 40 deletions.
39 changes: 28 additions & 11 deletions ...4j-core/src/main/java/org/deeplearning4j/datasets/canova/RecordReaderDataSetIterator.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -25,14 +25,18 @@
import org.canova.api.records.reader.RecordReader;
import org.canova.api.records.reader.SequenceRecordReader;
import org.canova.api.writable.Writable;
import org.canova.common.data.NDArrayWritable;
import org.deeplearning4j.datasets.iterator.DataSetIterator;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.dataset.DataSet;
import org.nd4j.linalg.dataset.api.DataSetPreProcessor;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.util.FeatureUtil;

import java.util.*;
import java.util.ArrayList;
import java.util.Collection;
import java.util.Iterator;
import java.util.List;


/**
Expand Down Expand Up @@ -204,15 +208,7 @@ private DataSet getDataSet(Collection<Writable> record) {
}

INDArray label = null;
INDArray featureVector;
if(regression && labelIndex >= 0){
//Handle the possibly multi-label regression case here:
int nLabels = labelIndexTo - labelIndex + 1;
featureVector = Nd4j.create(1, currList.size() - nLabels);
} else {
//Classification case, and also no-labels case
featureVector = Nd4j.create(labelIndex >= 0 ? currList.size() - 1 : currList.size());
}
INDArray featureVector = null;
int featureCount = 0;
int labelCount = 0;
for (int j = 0; j < currList.size(); j++) {
Expand Down Expand Up @@ -243,7 +239,28 @@ private DataSet getDataSet(Collection<Writable> record) {
label = FeatureUtil.toOutcomeVector(curr, numPossibleLabels);
}
} else {
featureVector.putScalar(featureCount++, current.toDouble());
try {
double value = current.toDouble();
if (featureVector == null) {
if(regression && labelIndex >= 0){
//Handle the possibly multi-label regression case here:
int nLabels = labelIndexTo - labelIndex + 1;
featureVector = Nd4j.create(1, currList.size() - nLabels);
} else {
//Classification case, and also no-labels case
featureVector = Nd4j.create(labelIndex >= 0 ? currList.size() - 1 : currList.size());
}
}
featureVector.putScalar(featureCount++, value);
} catch (UnsupportedOperationException e) {
// This isn't a scalar, so check if we got an array already
if (current instanceof NDArrayWritable) {
assert featureVector == null;
featureVector = ((NDArrayWritable)current).get();
} else {
throw e;
}
}
}
}

Expand Down
61 changes: 56 additions & 5 deletions ...re/src/main/java/org/deeplearning4j/datasets/canova/RecordReaderMultiDataSetIterator.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -22,14 +22,22 @@
import org.canova.api.records.reader.RecordReader;
import org.canova.api.records.reader.SequenceRecordReader;
import org.canova.api.writable.Writable;
import org.canova.common.data.NDArrayWritable;
import org.deeplearning4j.berkeley.Pair;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.dataset.api.MultiDataSet;
import org.nd4j.linalg.dataset.api.MultiDataSetPreProcessor;
import org.nd4j.linalg.dataset.api.iterator.MultiDataSetIterator;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.indexing.NDArrayIndex;

import java.util.*;
import java.util.ArrayList;
import java.util.Collection;
import java.util.HashMap;
import java.util.Iterator;
import java.util.List;
import java.util.Map;
import java.util.NoSuchElementException;

/**RecordReaderMultiDataSetIterator: A {@link MultiDataSetIterator} for data from one or more RecordReaders and SequenceRecordReaders<br>
* The idea: generate multiple inputs and multiple outputs from one or more Sequence/RecordReaders. Inputs and outputs
Expand Down Expand Up @@ -204,7 +212,16 @@ private INDArray convertWritables(List<Collection<Writable>> list, int minValues
int j = 0;
for (Writable w : c) {
idx[1] = j++;
arr.putScalar(idx, w.toDouble());
try {
arr.putScalar(idx, w.toDouble());
} catch (UnsupportedOperationException e) {
// This isn't a scalar, so check if we got an array already
if (w instanceof NDArrayWritable) {
arr.putRow(idx[0], ((NDArrayWritable)w).get());
} else {
throw e;
}
}
}
} else if(details.oneHot){
//Convert a single column to a one-hot representation
Expand All @@ -223,7 +240,18 @@ private INDArray convertWritables(List<Collection<Writable>> list, int minValues
int k=0;
for( int j=details.subsetStart; j<=details.subsetEndInclusive; j++){
idx[1] = k++;
arr.putScalar(idx,iter.next().toDouble());
Writable w = iter.next();
try {
arr.putScalar(idx,w.toDouble());
} catch (UnsupportedOperationException e) {
// This isn't a scalar, so check if we got an array already
if (w instanceof NDArrayWritable) {
arr.putRow(idx[0], ((NDArrayWritable)w).get().get(NDArrayIndex.all(),
NDArrayIndex.interval(details.subsetStart, details.subsetEndInclusive + 1)));
} else {
throw e;
}
}
}
}
}
Expand Down Expand Up @@ -278,7 +306,18 @@ private Pair<INDArray,INDArray> convertWritablesSequence(List<Collection<Collect
int j = 0;
while (iter.hasNext()) {
idx[1] = j++;
arr.putScalar(idx,iter.next().toDouble());
Writable w = iter.next();
try {
arr.putScalar(idx,w.toDouble());
} catch (UnsupportedOperationException e) {
// This isn't a scalar, so check if we got an array already
if (w instanceof NDArrayWritable) {
arr.get(NDArrayIndex.point(idx[0]), NDArrayIndex.all(), NDArrayIndex.point(idx[2]))
.putRow(0, ((NDArrayWritable)w).get());
} else {
throw e;
}
}
}
} else if(details.oneHot){
//Convert a single column to a one-hot representation
Expand All @@ -298,7 +337,19 @@ private Pair<INDArray,INDArray> convertWritablesSequence(List<Collection<Collect
int k=0;
for( int j=details.subsetStart; j<=details.subsetEndInclusive; j++){
idx[1] = k++;
arr.putScalar(idx,iter.next().toDouble());
Writable w = iter.next();
try {
arr.putScalar(idx,w.toDouble());
} catch (UnsupportedOperationException e) {
// This isn't a scalar, so check if we got an array already
if (w instanceof NDArrayWritable) {
arr.get(NDArrayIndex.point(idx[0]), NDArrayIndex.all(), NDArrayIndex.point(idx[2]))
.putRow(0, ((NDArrayWritable)w).get().get(NDArrayIndex.all(),
NDArrayIndex.interval(details.subsetStart, details.subsetEndInclusive + 1)));
} else {
throw e;
}
}
}
}
}
Expand Down
29 changes: 26 additions & 3 deletions ...src/main/java/org/deeplearning4j/datasets/canova/SequenceRecordReaderDataSetIterator.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@

import org.canova.api.records.reader.SequenceRecordReader;
import org.canova.api.writable.Writable;
import org.canova.common.data.NDArrayWritable;
import org.deeplearning4j.datasets.iterator.DataSetIterator;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.dataset.DataSet;
Expand All @@ -11,7 +12,11 @@
import org.nd4j.linalg.indexing.NDArrayIndex;
import org.nd4j.linalg.util.FeatureUtil;

import java.util.*;
import java.util.ArrayList;
import java.util.Collection;
import java.util.Iterator;
import java.util.List;
import java.util.NoSuchElementException;

/**
* Sequence record reader data set iterator
Expand Down Expand Up @@ -438,7 +443,16 @@ private INDArray getFeatures(Collection<Collection<Writable>> features) {
int f = 0;
while (timeStepIter.hasNext()) {
Writable current = timeStepIter.next();
out.put(i, f++, current.toDouble());
try {
out.put(i, f++, current.toDouble());
} catch (UnsupportedOperationException e) {
// This isn't a scalar, so check if we got an array already
if (current instanceof NDArrayWritable) {
out.putRow(i, ((NDArrayWritable)current).get());
} else {
throw e;
}
}
}
i++;
}
Expand Down Expand Up @@ -514,7 +528,16 @@ private INDArray[] getFeaturesLabelsSingleReader(Collection<Collection<Writable>
}
} else {
//feature
features.put(i, countFeatures++, current.toDouble());
try {
features.put(i, countFeatures++, current.toDouble());
} catch (UnsupportedOperationException e) {
// This isn't a scalar, so check if we got an array already
if (current instanceof NDArrayWritable) {
features.putRow(i, ((NDArrayWritable)current).get());
} else {
throw e;
}
}
}
}
i++;
Expand Down
12 changes: 6 additions & 6 deletions ...4j-core/src/main/java/org/deeplearning4j/datasets/iterator/impl/CifarDataSetIterator.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,7 @@ public class CifarDataSetIterator extends RecordReaderDataSetIterator {
* @param numExamples the overall number of examples
* */
public CifarDataSetIterator(int batchSize, int numExamples) {
super(null, batchSize, height * width * channels , CifarLoader.NUM_LABELS);
super(null, batchSize, 1 , CifarLoader.NUM_LABELS);
this.loader = new CifarLoader();
this.inputStream = loader.getInputStream();
this.numExamples = numExamples > totalExamples? totalExamples: numExamples;
Expand All @@ -47,7 +47,7 @@ public CifarDataSetIterator(int batchSize, int numExamples) {
* @param numExamples the overall number of examples
* */
public CifarDataSetIterator(int batchSize, int numExamples, String version) {
super(null, batchSize, height * width * channels, CifarLoader.NUM_LABELS);
super(null, batchSize, 1, CifarLoader.NUM_LABELS);
this.loader = new CifarLoader(version);
this.inputStream = loader.getInputStream();
this.numExamples = numExamples > totalExamples? totalExamples: numExamples;
Expand All @@ -61,7 +61,7 @@ public CifarDataSetIterator(int batchSize, int numExamples, String version) {
* @param numExamples the overall number of examples
* */
public CifarDataSetIterator(int batchSize, int numExamples, int[] imgDim) {
super(null, batchSize, imgDim[0] * imgDim[1] * imgDim[2], CifarLoader.NUM_LABELS);
super(null, batchSize, 1, CifarLoader.NUM_LABELS);
this.loader = new CifarLoader();
this.inputStream = loader.getInputStream();
this.numExamples = numExamples > totalExamples? totalExamples: numExamples;
Expand All @@ -74,7 +74,7 @@ public CifarDataSetIterator(int batchSize, int numExamples, int[] imgDim) {
* @param numExamples the overall number of examples
* */
public CifarDataSetIterator(int batchSize, int numExamples, int[] imgDim, int numCategories) {
super(null, batchSize, imgDim[0] * imgDim[1] * imgDim[2], numCategories);
super(null, batchSize, 1, numCategories);
this.loader = new CifarLoader();
this.inputStream = loader.getInputStream();
this.numExamples = numExamples > totalExamples? totalExamples: numExamples;
Expand All @@ -87,7 +87,7 @@ public CifarDataSetIterator(int batchSize, int numExamples, int[] imgDim, int nu
* @param numCategories the overall number of labels
* */
public CifarDataSetIterator(int batchSize, int numExamples, int numCategories) {
super(null, batchSize, height * width * channels, numCategories);
super(null, batchSize, 1, numCategories);
this.loader = new CifarLoader();
this.inputStream = loader.getInputStream();
this.numExamples = numExamples > totalExamples? totalExamples: numExamples;
Expand All @@ -100,7 +100,7 @@ public CifarDataSetIterator(int batchSize, int numExamples, int numCategories) {
* @param imgDim an array of height, width and channels
*/
public CifarDataSetIterator(int batchSize, int[] imgDim) {
super(null, batchSize, imgDim[0] * imgDim[1] * imgDim[2], CifarLoader.NUM_LABELS);
super(null, batchSize, 1, CifarLoader.NUM_LABELS);
this.loader = new CifarLoader();
this.inputStream = loader.getInputStream();
}
Expand Down
16 changes: 8 additions & 8 deletions ...ng4j-core/src/main/java/org/deeplearning4j/datasets/iterator/impl/LFWDataSetIterator.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -36,7 +36,7 @@ public class LFWDataSetIterator extends RecordReaderDataSetIterator {
* @param numExamples the overall number of examples
* */
public LFWDataSetIterator(int batchSize, int numExamples) {
super(new LFWLoader().getRecordReader(numExamples), batchSize, height * width * channels, LFWLoader.NUM_LABELS);
super(new LFWLoader().getRecordReader(numExamples), batchSize, 1, LFWLoader.NUM_LABELS);
}

/**
Expand All @@ -46,7 +46,7 @@ public LFWDataSetIterator(int batchSize, int numExamples) {
* @param numExamples the overall number of examples
* */
public LFWDataSetIterator(int batchSize, int numExamples, int[] imgDim) {
super(new LFWLoader().getRecordReader(numExamples, imgDim[0], imgDim[1], imgDim[2]), batchSize, imgDim[0] * imgDim[1] * imgDim[2], LFWLoader.NUM_LABELS);
super(new LFWLoader().getRecordReader(numExamples, imgDim[0], imgDim[1], imgDim[2]), batchSize, 1, LFWLoader.NUM_LABELS);
}

/**
Expand All @@ -56,7 +56,7 @@ public LFWDataSetIterator(int batchSize, int numExamples, int[] imgDim) {
* @param numExamples the overall number of examples
* */
public LFWDataSetIterator(int batchSize, int numExamples, int[] imgDim, int numCategories) {
super(new LFWLoader().getRecordReader(numExamples, imgDim[0], imgDim[1], imgDim[2]), batchSize, imgDim[0] * imgDim[1] * imgDim[2], numCategories);
super(new LFWLoader().getRecordReader(numExamples, imgDim[0], imgDim[1], imgDim[2]), batchSize, 1, numCategories);
}

/**
Expand All @@ -66,7 +66,7 @@ public LFWDataSetIterator(int batchSize, int numExamples, int[] imgDim, int numC
* @param numCategories the overall number of labels
* */
public LFWDataSetIterator(int batchSize, int numExamples, int numCategories) {
super(new LFWLoader().getRecordReader(numExamples, numCategories), batchSize, height * width * channels, numCategories);
super(new LFWLoader().getRecordReader(numExamples, numCategories), batchSize, 1, numCategories);
}

/**
Expand All @@ -75,7 +75,7 @@ public LFWDataSetIterator(int batchSize, int numExamples, int numCategories) {
* @param imgDim an array of height, width and channels
*/
public LFWDataSetIterator(int batchSize, int[] imgDim) {
super(new LFWLoader().getRecordReader(imgDim[0], imgDim[1], imgDim[2]), batchSize, imgDim[0] * imgDim[1] * imgDim[2], LFWLoader.NUM_LABELS);
super(new LFWLoader().getRecordReader(imgDim[0], imgDim[1], imgDim[2]), batchSize, 1, LFWLoader.NUM_LABELS);
}


Expand All @@ -86,7 +86,7 @@ public LFWDataSetIterator(int batchSize, int[] imgDim) {
* @param numCategories the overall number of labels
* */
public LFWDataSetIterator(int batchSize, int numExamples, int numCategories, boolean useSubset) {
super(new LFWLoader(useSubset).getRecordReader(numExamples, numCategories), batchSize, height * width * channels, numCategories);
super(new LFWLoader(useSubset).getRecordReader(numExamples, numCategories), batchSize, 1, numCategories);
}

/**
Expand All @@ -95,7 +95,7 @@ public LFWDataSetIterator(int batchSize, int numExamples, int numCategories, boo
* @param imgDim an array of height, width and channels
*/
public LFWDataSetIterator(int batchSize, int[] imgDim, boolean useSubset) {
super(new LFWLoader(useSubset).getRecordReader(imgDim[0], imgDim[1], imgDim[2]), batchSize, imgDim[0] * imgDim[1] * imgDim[2], useSubset ? LFWLoader.SUB_NUM_LABELS : LFWLoader.NUM_LABELS);
super(new LFWLoader(useSubset).getRecordReader(imgDim[0], imgDim[1], imgDim[2]), batchSize, 1, useSubset ? LFWLoader.SUB_NUM_LABELS : LFWLoader.NUM_LABELS);
}


Expand All @@ -106,7 +106,7 @@ public LFWDataSetIterator(int batchSize, int[] imgDim, boolean useSubset) {
* @param numExamples the overall number of examples
* */
public LFWDataSetIterator(int batchSize, int numExamples, int[] imgDim, int numCategories, boolean useSubset, Random rng) {
super(new LFWLoader(useSubset).getRecordReader(numExamples, imgDim[0], imgDim[1], imgDim[2], numCategories, rng), batchSize, imgDim[0] * imgDim[1] * imgDim[2], numCategories);
super(new LFWLoader(useSubset).getRecordReader(numExamples, imgDim[0], imgDim[1], imgDim[2], numCategories, rng), batchSize, 1, numCategories);
}

}
4 changes: 2 additions & 2 deletions ...ore/src/test/java/org/deeplearning4j/nn/layers/convolution/ConvolutionLayerSetupTest.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -132,7 +132,7 @@ public void testMultiChannel() throws Exception {

RecordReader reader = new ImageRecordReader(28,28,3,true,labels);
reader.initialize(new FileSplit(new File(rootDir)));
DataSetIterator recordReader = new RecordReaderDataSetIterator(reader,28 * 28 * 3,labels.size());
DataSetIterator recordReader = new RecordReaderDataSetIterator(reader,1,labels.size());

labels.remove("lfwtest");
NeuralNetConfiguration.ListBuilder builder = (NeuralNetConfiguration.ListBuilder) incompleteLFW();
Expand All @@ -154,7 +154,7 @@ public void testLRN() throws Exception{

RecordReader reader = new ImageRecordReader(28,28,3,true,labels);
reader.initialize(new FileSplit(new File(rootDir)));
DataSetIterator recordReader = new RecordReaderDataSetIterator(reader,28 * 28 * 3,labels.size());
DataSetIterator recordReader = new RecordReaderDataSetIterator(reader,1,labels.size());
labels.remove("lfwtest");
NeuralNetConfiguration.ListBuilder builder = (NeuralNetConfiguration.ListBuilder) incompleteLRN();
new ConvolutionLayerSetup(builder,28,28,3);
Expand Down
Loading

0 comments on commit 6a0efff

Please sign in to comment.