Brutex
/
cavis
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

DL4J/ND4J: Do pass on integer casts (#15) 

* Int cast fixes.

* Revert "Int cast fixes."

This reverts commit aa36e8ca

* Int casts

* Int cast

* Int casts

* Get rid of int casts. Dropping deprecated aggregate ops.

* java scatterUpdate changes

Signed-off-by: raver119 <raver119@gmail.com>

* c++ scatterUpdate changes

Signed-off-by: raver119 <raver119@gmail.com>

* Remove aggregated ops.

* Restored test

* Tests restored.

* Minor fixes
master
Alexander Stoyakin 2019-10-31 11:23:09 +02:00 committed by GitHub
parent 95f7ad7b94
commit 45a40c8a89
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
177 changed files with 704 additions and 1988 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
arbiter/arbiter-deeplearning4j/src/test/java/org/deeplearning4j/arbiter/TestUtils.java
View File

@ -124,7 +124,6 @@ public class TestUtils {
public static INDArray randomOneHot(long examples, long nOut, Random rng){
INDArray arr = Nd4j.create(examples, nOut);
for( int i=0; i<examples; i++ ){
// FIXME: int cast
arr.putScalar(i, rng.nextInt((int) nOut), 1.0);
}
return arr;

1
deeplearning4j/deeplearning4j-core/src/test/java/org/deeplearning4j/TestUtils.java
View File

@ -132,7 +132,6 @@ public class TestUtils {
public static INDArray randomOneHot(long examples, long nOut, Random rng){
INDArray arr = Nd4j.create(examples, nOut);
for( int i=0; i<examples; i++ ){
// FIXME: int cast
arr.putScalar(i, rng.nextInt((int) nOut), 1.0);
}
return arr;

1
deeplearning4j/deeplearning4j-core/src/test/java/org/deeplearning4j/datasets/iterator/AsyncDataSetIteratorTest.java
View File

@ -187,7 +187,6 @@ public class AsyncDataSetIteratorTest extends BaseDL4JTest {
}
}
@Test
public void testVariableTimeSeries2() throws Exception {
AsyncDataSetIterator adsi =

4
deeplearning4j/deeplearning4j-core/src/test/java/org/deeplearning4j/gradientcheck/LossFunctionGradientCheck.java
View File

@ -464,13 +464,11 @@ public class LossFunctionGradientCheck extends BaseDL4JTest {
ret[1] = Nd4j.zeros(labelsShape);
if (labelsShape.length == 2) {
for (int i = 0; i < labelsShape[0]; i++) {
// FIXME: int cast
ret[1].putScalar(i, r.nextInt((int) labelsShape[1]), 1.0);
}
} else if (labelsShape.length == 3) {
for (int i = 0; i < labelsShape[0]; i++) {
for (int j = 0; j < labelsShape[2]; j++) {
// FIXME: int cast
ret[1].putScalar(i, r.nextInt((int) labelsShape[1]), j, 1.0);
}
}
@ -484,13 +482,11 @@ public class LossFunctionGradientCheck extends BaseDL4JTest {
ret[1] = Nd4j.ones(labelsShape);
if (labelsShape.length == 2) {
for (int i = 0; i < labelsShape[0]; i++) {
// FIXME: int cast
ret[1].putScalar(i, r.nextInt((int) labelsShape[1]), -1.0);
}
} else if (labelsShape.length == 3) {
for (int i = 0; i < labelsShape[0]; i++) {
for (int j = 0; j < labelsShape[2]; j++) {
// FIXME: int cast
ret[1].putScalar(i, r.nextInt((int) labelsShape[1]), j, -1.0);
}
}

3
deeplearning4j/deeplearning4j-core/src/test/java/org/deeplearning4j/nn/conf/graph/ShiftVertexTest.java
View File

@ -176,8 +176,7 @@ public class ShiftVertexTest extends BaseDL4JTest {
manual_weights.put("output_b", c);
// First things first, let's calculate the score.
// FIXME: int cast
int batchsz = (int) input.shape()[0];
long batchsz = input.shape()[0];
INDArray z = input.castTo(W.dataType()).mmul(W).add(b.repmat(batchsz, 1));
INDArray a = a1.getActivation(z.dup(), true).add(sf); // activation modifies it's input!!
INDArray q = a.mmul(V).add(c.repmat(batchsz, 1));

4
deeplearning4j/deeplearning4j-core/src/test/java/org/deeplearning4j/nn/multilayer/BackPropMLPTest.java
View File

@ -36,6 +36,7 @@ import org.nd4j.linalg.api.ops.impl.transforms.strict.SigmoidDerivative;
import org.nd4j.linalg.api.ops.impl.transforms.strict.TanhDerivative;
import org.nd4j.linalg.dataset.DataSet;
import org.nd4j.linalg.dataset.api.iterator.DataSetIterator;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.learning.config.Sgd;
import org.nd4j.linalg.lossfunctions.LossFunctions.LossFunction;
@ -340,7 +341,8 @@ public class BackPropMLPTest extends BaseDL4JTest {
public static float[] asFloat(INDArray arr) {
long len = arr.length();
// FIXME: int cast
if (len > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
float[] f = new float[(int) len];
NdIndexIterator iterator = new NdIndexIterator('c', arr.shape());
for (int i = 0; i < len; i++) {

1
deeplearning4j/deeplearning4j-core/src/test/java/org/deeplearning4j/nn/multilayer/MultiLayerTest.java
View File

@ -320,7 +320,6 @@ public class MultiLayerTest extends BaseDL4JTest {
public static float[] asFloat(INDArray arr) {
long len = arr.length();
// FIXME: int cast
float[] f = new float[(int) len];
for (int i = 0; i < len; i++)
f[i] = arr.getFloat(i);

1
deeplearning4j/deeplearning4j-core/src/test/java/org/deeplearning4j/nn/updater/TestUpdaters.java
View File

@ -331,7 +331,6 @@ public class TestUpdaters extends BaseDL4JTest {
double calculatedByHandMScalar = 0.2;
double[] expectedM = Nd4j.ones(1, numParams).mul(calculatedByHandMScalar).data().asDouble();
// FIXME: int cast
double[] actualM = Arrays.copyOfRange(nadamUpdater.getM().data().asDouble(), 0, (int) numParams);
for (int i = 0; i < actualM.length; i++) {
actualM[i] = Math.round(actualM[i] * 1e2) / 1e2;

8
deeplearning4j/deeplearning4j-core/src/test/java/org/deeplearning4j/optimize/solver/TestOptimizers.java
View File

@ -48,6 +48,7 @@ import org.nd4j.linalg.api.rng.DefaultRandom;
import org.nd4j.linalg.api.rng.Random;
import org.nd4j.linalg.dataset.DataSet;
import org.nd4j.linalg.dataset.api.iterator.DataSetIterator;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.indexing.conditions.Condition;
import org.nd4j.linalg.learning.config.AdaGrad;
@ -664,7 +665,9 @@ public class TestOptimizers extends BaseDL4JTest {
double xlm1 = parameters.getDouble(nDims - 2);
double gl = 200 * (xl - xlm1 * xlm1);
// FIXME: int cast
if (nDims - 1 > Integer.MAX_VALUE) {
throw new ND4JArraySizeException();
}
gradient.put(0, (int)nDims - 1, gl);
Gradient g = new DefaultGradient();
g.gradientForVariable().put("W", gradient);
@ -865,8 +868,7 @@ public class TestOptimizers extends BaseDL4JTest {
@Override
public long numParams() {
// FIXME: int cast
return (int) parameters.length();
return parameters.length();
}
@Override

2
deeplearning4j/deeplearning4j-data/deeplearning4j-datavec-iterators/src/main/java/org/deeplearning4j/datasets/datavec/RecordReaderMultiDataSetIterator.java
View File

@ -286,7 +286,7 @@ public class RecordReaderMultiDataSetIterator implements MultiDataSetIterator, S
for (INDArray w : exampleData) {
val n = w.size(0);
// FIXME: int cast
if (Math.min(minExamples, n) < Integer.MAX_VALUE)
minExamples = (int) Math.min(minExamples, n);
}
}

2
deeplearning4j/deeplearning4j-data/deeplearning4j-datavec-iterators/src/main/java/org/deeplearning4j/datasets/datavec/SequenceRecordReaderDataSetIterator.java
View File

@ -366,7 +366,6 @@ public class SequenceRecordReaderDataSetIterator implements DataSetIterator {
DataSet ds = mdsToDataSet(mds);
if (totalOutcomes == -1) {
// FIXME: int cast
inputColumns = (int) ds.getFeatures().size(1);
totalOutcomes = ds.getLabels() == null ? -1 : (int) ds.getLabels().size(1);
}
@ -394,7 +393,6 @@ public class SequenceRecordReaderDataSetIterator implements DataSetIterator {
stored = next();
useStored = true;
// FIXME: int cast
inputColumns = (int) stored.getFeatures().size(1);
totalOutcomes = (int) stored.getLabels().size(1);
}

1
deeplearning4j/deeplearning4j-data/deeplearning4j-utility-iterators/src/main/java/org/deeplearning4j/datasets/iterator/AbstractDataSetIterator.java
View File

@ -172,7 +172,6 @@ public abstract class AbstractDataSetIterator<T> implements DataSetIterator {
Pair<T, T> pair = iterator.next();
if (numFeatures < 1) {
if (pair.getFirst() instanceof INDArray) {
// FIXME: int cast
numFeatures = (int) ((INDArray) pair.getFirst()).length();
numLabels = (int) ((INDArray) pair.getSecond()).length();
} else if (pair.getFirst() instanceof float[]) {

1
deeplearning4j/deeplearning4j-data/deeplearning4j-utility-iterators/src/main/java/org/deeplearning4j/datasets/iterator/IteratorDataSetIterator.java
View File

@ -95,7 +95,6 @@ public class IteratorDataSetIterator implements DataSetIterator {
//Set columns etc for later use
DataSet temp = list.get(0);
// FIXME: int cast
inputColumns = (int) temp.getFeatures().size(1);
totalOutcomes = temp.getLabels() == null ? 0 : (int) temp.getLabels().size(1); //May be null for layerwise pretraining
}

5
deeplearning4j/deeplearning4j-data/deeplearning4j-utility-iterators/src/main/java/org/deeplearning4j/datasets/iterator/IteratorMultiDataSetIterator.java
View File

@ -73,8 +73,7 @@ public class IteratorMultiDataSetIterator implements MultiDataSetIterator {
next = iterator.next();
}
// FIXME: int cast
int nExamples = (int) next.getFeatures(0).size(0);
long nExamples = next.getFeatures(0).size(0);
if (countSoFar + nExamples <= batchSize) {
//Add the entire MultiDataSet as-is
list.add(next);
@ -140,7 +139,7 @@ public class IteratorMultiDataSetIterator implements MultiDataSetIterator {
return out;
}
private static INDArray getRange(INDArray arr, int exampleFrom, int exampleToExclusive) {
private static INDArray getRange(INDArray arr, long exampleFrom, long exampleToExclusive) {
if (arr == null)
return null;

4
deeplearning4j/deeplearning4j-data/deeplearning4j-utility-iterators/src/main/java/org/deeplearning4j/datasets/iterator/file/BaseFileIterator.java
View File

@ -134,7 +134,7 @@ public abstract class BaseFileIterator<T, P> implements Iterator<T> {
List<T> remainder = new ArrayList<>();
int soFar = 0;
for (T t : toMerge) {
int size = sizeOf(t);
long size = sizeOf(t);
if (soFar + size <= batchSize) {
correctNum.add(t);
@ -190,7 +190,7 @@ public abstract class BaseFileIterator<T, P> implements Iterator<T> {
protected abstract T load(File f);
protected abstract int sizeOf(T of);
protected abstract long sizeOf(T of);
protected abstract List<T> split(T toSplit);

2
deeplearning4j/deeplearning4j-data/deeplearning4j-utility-iterators/src/main/java/org/deeplearning4j/datasets/iterator/file/FileDataSetIterator.java
View File

@ -151,7 +151,7 @@ public class FileDataSetIterator extends BaseFileIterator<DataSet, DataSetPrePro
}
@Override
protected int sizeOf(DataSet of) {
protected long sizeOf(DataSet of) {
return of.numExamples();
}

5
deeplearning4j/deeplearning4j-data/deeplearning4j-utility-iterators/src/main/java/org/deeplearning4j/datasets/iterator/file/FileMultiDataSetIterator.java
View File

@ -151,9 +151,8 @@ public class FileMultiDataSetIterator extends BaseFileIterator<MultiDataSet, Mul
}
@Override
protected int sizeOf(MultiDataSet of) {
// FIXME: int cast
return (int) of.getFeatures(0).size(0);
protected long sizeOf(MultiDataSet of) {
return of.getFeatures(0).size(0);
}
@Override

3
deeplearning4j/deeplearning4j-manifold/deeplearning4j-tsne/src/main/java/org/deeplearning4j/plot/BarnesHutTsne.java
View File

@ -665,8 +665,7 @@ public class BarnesHutTsne implements Model {
if (useAdaGrad) {
if (adaGrad == null) {
// FIXME: int cast
adaGrad = new AdaGrad(ArrayUtil.toInts(gradient.shape()), learningRate);
adaGrad = new AdaGrad(gradient.shape(), learningRate);
adaGrad.setStateViewArray(Nd4j.zeros(gradient.shape()).reshape(1, gradChange.length()),
gradChange.shape(), gradient.ordering(), true);
}

12
deeplearning4j/deeplearning4j-modelimport/src/main/java/org/deeplearning4j/nn/modelimport/keras/layers/core/KerasReshape.java
View File

@ -18,6 +18,7 @@ package org.deeplearning4j.nn.modelimport.keras.layers.core;
import lombok.val;
import org.apache.commons.lang3.ArrayUtils;
import org.deeplearning4j.nn.conf.InputPreProcessor;
import org.deeplearning4j.nn.conf.inputs.InputType;
import org.deeplearning4j.nn.modelimport.keras.KerasLayer;
@ -51,6 +52,13 @@ public class KerasReshape extends KerasLayer {
this(layerConfig, true);
}
private long[] listToLongArray(List<Integer> list) {
long[] retVal = new long[list.size()];
for (int i = 0; i < list.size(); ++i) {
retVal[i] = list.get(i);
}
return retVal;
}
/**
* Constructor from parsed Keras layer configuration dictionary.
*
@ -67,9 +75,7 @@ public class KerasReshape extends KerasLayer {
if (innerConfig.containsKey(targetShape)) {
@SuppressWarnings("unchecked")
List<Integer> targetShapeList = (List<Integer>) innerConfig.get(targetShape);
// FIXME: int cast
this.targetShape = ArrayUtil.toLongArray(ArrayUtil.toArray(targetShapeList));
this.targetShape = listToLongArray(targetShapeList);
}
}

2
deeplearning4j/deeplearning4j-modelimport/src/test/java/org/deeplearning4j/nn/modelimport/keras/e2e/KerasModelEndToEndTest.java
View File

@ -690,13 +690,11 @@ public class KerasModelEndToEndTest extends BaseDL4JTest {
INDArray testLabels = Nd4j.create(predictionsDl4j.shape());
if (testLabels.rank() == 2) {
for (int i = 0; i < testLabels.size(0); i++) {
// FIXME: int cast
testLabels.putScalar(i, r.nextInt((int) testLabels.size(1)), 1.0);
}
} else if (testLabels.rank() == 3) {
for (int i = 0; i < testLabels.size(0); i++) {
for (int j = 0; j < testLabels.size(1); j++) {
// FIXME: int cast
testLabels.putScalar(i, j, r.nextInt((int) testLabels.size(1)), 1.0);
}
}

6
deeplearning4j/deeplearning4j-nlp-parent/deeplearning4j-nlp/src/main/java/org/deeplearning4j/models/embeddings/inmemory/InMemoryLookupTable.java
View File

@ -104,7 +104,7 @@ public class InMemoryLookupTable<T extends SequenceElement> implements WeightLoo
}
protected void initAdaGrad() {
int[] shape = new int[] {vocab.numWords() + 1, vectorLength};
long[] shape = new long[] {vocab.numWords() + 1, vectorLength};
int length = ArrayUtil.prod(shape);
adaGrad = new AdaGrad(shape, lr.get());
adaGrad.setStateViewArray(Nd4j.zeros(shape).reshape(1, length), shape, Nd4j.order(), true);
@ -124,8 +124,7 @@ public class InMemoryLookupTable<T extends SequenceElement> implements WeightLoo
if (adaGrad == null)
initAdaGrad();
// FIXME: int cast
return adaGrad.getGradient(gradient, column, ArrayUtil.toInts(syn0.shape()));
return adaGrad.getGradient(gradient, column, syn0.shape());
}
@Override
@ -370,7 +369,6 @@ public class InMemoryLookupTable<T extends SequenceElement> implements WeightLoo
else {
nextRandom.set(nextRandom.get() * 25214903917L + 11);
// FIXME: int cast
int idx = (int) Math.abs((int) (nextRandom.get() >> 16) % table.length());
target = table.getInt(idx);

1
deeplearning4j/deeplearning4j-nlp-parent/deeplearning4j-nlp/src/main/java/org/deeplearning4j/models/embeddings/learning/impl/elements/CBOW.java
View File

@ -33,7 +33,6 @@ import org.deeplearning4j.models.word2vec.wordstore.VocabCache;
import org.nd4j.linalg.api.buffer.DataType;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.api.ops.aggregates.Aggregate;
import org.nd4j.linalg.api.ops.aggregates.impl.AggregateCBOW;
import org.nd4j.linalg.api.ops.impl.nlp.CbowRound;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.util.DeviceLocalNDArray;

11
deeplearning4j/deeplearning4j-nlp-parent/deeplearning4j-nlp/src/main/java/org/deeplearning4j/models/embeddings/learning/impl/elements/GloVe.java
View File

@ -104,11 +104,10 @@ public class GloVe<T extends SequenceElement> implements ElementsLearningAlgorit
weightAdaGrad = new AdaGrad(new int[] {this.vocabCache.numWords() + 1, vectorLength}, learningRate);
weightAdaGrad = new AdaGrad(new long[] {this.vocabCache.numWords() + 1, vectorLength}, learningRate);
bias = Nd4j.create(syn0.rows());
// FIXME: int cast
biasAdaGrad = new AdaGrad(ArrayUtil.toInts(bias.shape()), this.learningRate);
biasAdaGrad = new AdaGrad(bias.shape(), this.learningRate);
// maxmemory = Runtime.getRuntime().maxMemory() - (vocabCache.numWords() * vectorLength * 2 * 8);
@ -237,15 +236,13 @@ public class GloVe<T extends SequenceElement> implements ElementsLearningAlgorit
private void update(T element1, INDArray wordVector, INDArray contextVector, double gradient) {
//gradient for word vectors
INDArray grad1 = contextVector.mul(gradient);
// FIXME: int cast
INDArray update = weightAdaGrad.getGradient(grad1, element1.getIndex(), ArrayUtil.toInts(syn0.shape()));
INDArray update = weightAdaGrad.getGradient(grad1, element1.getIndex(), syn0.shape());
//update vector
wordVector.subi(update);
double w1Bias = bias.getDouble(element1.getIndex());
// FIXME: int cast
double biasGradient = biasAdaGrad.getGradient(gradient, element1.getIndex(), ArrayUtil.toInts(bias.shape()));
double biasGradient = biasAdaGrad.getGradient(gradient, element1.getIndex(), bias.shape());
double update2 = w1Bias - biasGradient;
bias.putScalar(element1.getIndex(), update2);
}

1
deeplearning4j/deeplearning4j-nlp-parent/deeplearning4j-nlp/src/main/java/org/deeplearning4j/models/embeddings/reader/impl/BasicModelUtils.java
View File

@ -358,7 +358,6 @@ public class BasicModelUtils<T extends SequenceElement> implements ModelUtils<T>
INDArray sort = sorted[0];
List<String> ret = new ArrayList<>();
// FIXME: int cast
if (top > sort.length())
top = (int) sort.length();
//there will be a redundant word

8
deeplearning4j/deeplearning4j-nlp-parent/deeplearning4j-nlp/src/main/java/org/deeplearning4j/models/glove/GloveWeightLookupTable.java
View File

@ -72,7 +72,7 @@ public class GloveWeightLookupTable<T extends SequenceElement> extends InMemoryL
putVector(Word2Vec.DEFAULT_UNK, randUnk);
}
if (weightAdaGrad == null || reset) {
weightAdaGrad = new AdaGrad(new int[] {vocab.numWords() + 1, vectorLength}, lr.get());
weightAdaGrad = new AdaGrad(new long[]{vocab.numWords() + 1, vectorLength}, lr.get());
}
@ -81,7 +81,7 @@ public class GloveWeightLookupTable<T extends SequenceElement> extends InMemoryL
bias = Nd4j.create(syn0.rows());
if (biasAdaGrad == null || reset) {
biasAdaGrad = new AdaGrad(ArrayUtil.toInts(bias.shape()), lr.get());
biasAdaGrad = new AdaGrad(bias.shape(), lr.get());
}
@ -140,13 +140,13 @@ public class GloveWeightLookupTable<T extends SequenceElement> extends InMemoryL
private void update(T w1, INDArray wordVector, INDArray contextVector, double gradient) {
//gradient for word vectors
INDArray grad1 = contextVector.mul(gradient);
INDArray update = weightAdaGrad.getGradient(grad1, w1.getIndex(), ArrayUtil.toInts(syn0.shape()));
INDArray update = weightAdaGrad.getGradient(grad1, w1.getIndex(), syn0.shape());
//update vector
wordVector.subi(update);
double w1Bias = bias.getDouble(w1.getIndex());
double biasGradient = biasAdaGrad.getGradient(gradient, w1.getIndex(), ArrayUtil.toInts(bias.shape()));
double biasGradient = biasAdaGrad.getGradient(gradient, w1.getIndex(), bias.shape());
double update2 = w1Bias - biasGradient;
bias.putScalar(w1.getIndex(), update2);
}

4
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/gradientcheck/GradientCheckUtil.java
View File

@ -20,6 +20,7 @@ import lombok.extern.slf4j.Slf4j;
import lombok.val;
import org.deeplearning4j.nn.api.Model;
import org.nd4j.linalg.api.buffer.DataType;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.function.Consumer;
import org.nd4j.linalg.lossfunctions.impl.LossBinaryXENT;
import org.nd4j.linalg.primitives.Pair;
@ -293,7 +294,8 @@ public class GradientCheckUtil {
ss = n;
}
// FIXME: int cast
if (ss > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
stepSizeForParam.put(paramNames.get(i), (int) ss);
}
}

14
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/conf/graph/ElementWiseVertex.java
View File

@ -140,10 +140,9 @@ public class ElementWiseVertex extends GraphVertex {
//CNN inputs... also check that the channels, width and heights match:
InputType.InputTypeConvolutional firstConv = (InputType.InputTypeConvolutional) first;
// FIXME: int cast
val fd = (int) firstConv.getChannels();
val fw = (int) firstConv.getWidth();
val fh = (int) firstConv.getHeight();
val fd = firstConv.getChannels();
val fw = firstConv.getWidth();
val fh = firstConv.getHeight();
for (int i = 1; i < vertexInputs.length; i++) {
if (vertexInputs[i].getType() != InputType.Type.CNN) {
@ -155,10 +154,9 @@ public class ElementWiseVertex extends GraphVertex {
InputType.InputTypeConvolutional otherConv = (InputType.InputTypeConvolutional) vertexInputs[i];
// FIXME: int cast
val od = (int) otherConv.getChannels();
val ow = (int) otherConv.getWidth();
val oh = (int) otherConv.getHeight();
val od = otherConv.getChannels();
val ow = otherConv.getWidth();
val oh = otherConv.getHeight();
if (fd != od || fw != ow || fh != oh) {
throw new InvalidInputTypeException(

35
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/conf/graph/MergeVertex.java
View File

@ -94,13 +94,12 @@ public class MergeVertex extends GraphVertex {
// CNN3D inputs: check that the channels, width and height match:
InputType.InputTypeConvolutional3D firstConv = (InputType.InputTypeConvolutional3D) first;
// FIXME: int cast
val fd = (int) firstConv.getDepth();
val fw = (int) firstConv.getWidth();
val fh = (int) firstConv.getHeight();
val fc = (int) firstConv.getChannels();
val fd = firstConv.getDepth();
val fw = firstConv.getWidth();
val fh = firstConv.getHeight();
val fc = firstConv.getChannels();
int depthSum = fc;
long depthSum = fc;
InputType.InputTypeConvolutional3D otherConv = null;
for (int i = 1; i < vertexInputs.length; i++) {
if (vertexInputs[i].getType() != InputType.Type.CNN3D) {
@ -109,10 +108,10 @@ public class MergeVertex extends GraphVertex {
}
otherConv = (InputType.InputTypeConvolutional3D) vertexInputs[i];
val od = (int) otherConv.getDepth();
val ow = (int) otherConv.getWidth();
val oh = (int) otherConv.getHeight();
val oc = (int) otherConv.getChannels();
val od = otherConv.getDepth();
val ow = otherConv.getWidth();
val oh = otherConv.getHeight();
val oc = otherConv.getChannels();
if (fd != od || fw != ow || fh != oh) {
throw new InvalidInputTypeException("Invalid input: MergeVertex cannot merge CNN3D activations of different width/heights:" + "first [channels,width,height] = [" + fd + "," + fw + "," + fh
@ -177,12 +176,11 @@ public class MergeVertex extends GraphVertex {
//CNN inputs... also check that the channels, width and heights match:
InputType.InputTypeConvolutional firstConv = (InputType.InputTypeConvolutional) first;
// FIXME: int cast
val fd = (int) firstConv.getChannels();
val fw = (int) firstConv.getWidth();
val fh = (int) firstConv.getHeight();
val fd = firstConv.getChannels();
val fw = firstConv.getWidth();
val fh = firstConv.getHeight();
int depthSum = fd;
long depthSum = fd;
for (int i = 1; i < vertexInputs.length; i++) {
if (vertexInputs[i].getType() != InputType.Type.CNN) {
@ -194,10 +192,9 @@ public class MergeVertex extends GraphVertex {
InputType.InputTypeConvolutional otherConv = (InputType.InputTypeConvolutional) vertexInputs[i];
// FIXME: int cast
val od = (int) otherConv.getChannels();
val ow = (int) otherConv.getWidth();
val oh = (int) otherConv.getHeight();
val od = otherConv.getChannels();
val ow = otherConv.getWidth();
val oh = otherConv.getHeight();
if (fw != ow || fh != oh) {
throw new InvalidInputTypeException(

16
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/conf/graph/PoolHelperVertex.java
View File

@ -131,12 +131,11 @@ public class PoolHelperVertex extends GraphVertex {
//CNN inputs... also check that the channels, width and heights match:
InputType.InputTypeConvolutional firstConv = (InputType.InputTypeConvolutional) first;
// FIXME: int cast
val fd = (int) firstConv.getChannels();
val fw = (int) firstConv.getWidth();
val fh = (int) firstConv.getHeight();
val fd = firstConv.getChannels();
val fw = firstConv.getWidth();
val fh = firstConv.getHeight();
int depthSum = fd;
long depthSum = fd;
for (int i = 1; i < vertexInputs.length; i++) {
if (vertexInputs[i].getType() != InputType.Type.CNN) {
@ -148,10 +147,9 @@ public class PoolHelperVertex extends GraphVertex {
InputType.InputTypeConvolutional otherConv = (InputType.InputTypeConvolutional) vertexInputs[i];
// FIXME: int cast
int od = (int) otherConv.getChannels();
int ow = (int) otherConv.getWidth();
int oh = (int) otherConv.getHeight();
long od = otherConv.getChannels();
long ow = otherConv.getWidth();
long oh = otherConv.getHeight();
if (fw != ow || fh != oh) {
throw new InvalidInputTypeException(

16
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/conf/graph/UnstackVertex.java
View File

@ -150,12 +150,11 @@ public class UnstackVertex extends GraphVertex {
//CNN inputs... also check that the channels, width and heights match:
InputType.InputTypeConvolutional firstConv = (InputType.InputTypeConvolutional) first;
// FIXME: int cast
val fd = (int) firstConv.getChannels();
val fw = (int) firstConv.getWidth();
val fh = (int) firstConv.getHeight();
val fd = firstConv.getChannels();
val fw = firstConv.getWidth();
val fh = firstConv.getHeight();
int depthSum = fd;
long depthSum = fd;
for (int i = 1; i < vertexInputs.length; i++) {
if (vertexInputs[i].getType() != InputType.Type.CNN) {
@ -167,10 +166,9 @@ public class UnstackVertex extends GraphVertex {
InputType.InputTypeConvolutional otherConv = (InputType.InputTypeConvolutional) vertexInputs[i];
// FIXME: int cast
val od = (int) otherConv.getChannels();
val ow = (int) otherConv.getWidth();
val oh = (int) otherConv.getHeight();
val od = otherConv.getChannels();
val ow = otherConv.getWidth();
val oh = otherConv.getHeight();
if (fw != ow || fh != oh) {
throw new InvalidInputTypeException(

9
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/conf/inputs/InputType.java
View File

@ -402,18 +402,17 @@ public abstract class InputType implements Serializable {
//Note: ConvolutionalFlat and FeedForward look identical... but either should work OK if using something
// like FeedForwardToCnnPreProcessor
// FIXME: int cast
switch (inputArray.rank()) {
case 2:
return InputType.feedForward((int) inputArray.size(1));
return InputType.feedForward(inputArray.size(1));
case 3:
return InputType.recurrent((int) inputArray.size(1), (int) inputArray.size(2));
return InputType.recurrent(inputArray.size(1), (int) inputArray.size(2));
case 4:
//Order: [minibatch, channels, height, width] -> [h, w, c]
return InputType.convolutional((int) inputArray.size(2), (int) inputArray.size(3), (int) inputArray.size(1));
return InputType.convolutional(inputArray.size(2), (int) inputArray.size(3), (int) inputArray.size(1));
case 5:
//Order: [minibatch, channels, depth, height, width] -> [d, h, w, c]
return InputType.convolutional3D((int) inputArray.size(2), (int) inputArray.size(3),
return InputType.convolutional3D(inputArray.size(2), (int) inputArray.size(3),
(int) inputArray.size(4), (int) inputArray.size(1));
default:
throw new IllegalArgumentException(

5
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/conf/layers/Cnn3DLossLayer.java
View File

@ -152,17 +152,18 @@ public class Cnn3DLossLayer extends FeedForwardLayer {
}
@Override
public void setNIn(int nIn){
public void setNIn(long nIn){
throw new UnsupportedOperationException(
"Cnn3DLossLayer has no parameters, thus nIn will always equal nOut.");
}
@Override
public void setNOut(int nOut){
public void setNOut(long nOut){
throw new UnsupportedOperationException(
"Cnn3DLossLayer has no parameters, thus nIn will always equal nOut.");
}
@Override
@SuppressWarnings("unchecked")
public Cnn3DLossLayer build() {

4
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/conf/layers/CnnLossLayer.java
View File

@ -145,13 +145,13 @@ public class CnnLossLayer extends FeedForwardLayer {
}
@Override
public void setNIn(int nIn){
public void setNIn(long nIn){
throw new UnsupportedOperationException(
"This layer has no parameters, thus nIn will always equal nOut.");
}
@Override
public void setNOut(int nOut){
public void setNOut(long nOut){
throw new UnsupportedOperationException(
"This layer has no parameters, thus nIn will always equal nOut.");
}

2
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/conf/layers/Convolution1DLayer.java
View File

@ -88,7 +88,7 @@ public class Convolution1DLayer extends ConvolutionLayer {
//Probably: user did InputType.recurrent(x) without specifying sequence length
outLength = -1;
} else {
outLength = Convolution1DUtils.getOutputSize((int) inputTsLength, kernelSize[0], stride[0], padding[0],
outLength = Convolution1DUtils.getOutputSize(inputTsLength, kernelSize[0], stride[0], padding[0],
convolutionMode, dilation[0]);
}
return InputType.recurrent(nOut, outLength);

7
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/conf/layers/FeedForwardLayer.java
View File

@ -117,14 +117,14 @@ public abstract class FeedForwardLayer extends BaseLayer {
* this is the input channels, otherwise is the previous layer size.
*
*/
protected int nIn = 0;
protected long nIn = 0;
/**
* Number of inputs for the layer (usually the size of the last layer). <br> Note that for Convolutional layers,
* this is the input channels, otherwise is the previous layer size.
*
*/
protected int nOut = 0;
protected long nOut = 0;
/**
* Number of inputs for the layer (usually the size of the last layer). <br> Note that for Convolutional layers,
@ -144,8 +144,7 @@ public abstract class FeedForwardLayer extends BaseLayer {
* @param nIn Number of inputs for the layer
*/
public T nIn(long nIn) {
// FIXME: int cast
this.setNIn((int) nIn);
this.setNIn(nIn);
return (T) this;
}

37
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/conf/layers/InputTypeUtil.java
View File

@ -41,12 +41,9 @@ public class InputTypeUtil {
Class<?> layerClass) {
InputType.InputTypeConvolutional i = (InputType.InputTypeConvolutional) inputType;
// FIXME: int cast
val hIn = (int) i.getHeight();
val wIn = (int) i.getWidth();
val hIn = i.getHeight();
val wIn = i.getWidth();
val inHeight = (int) i.getHeight();
val inWidth = (int) i.getWidth();
int padH = (padding == null ? 0 : padding[0]); //May be null for ConvolutionMode.Same
int padW = (padding == null ? 0 : padding[1]);
int kH = kernelSize[0];
@ -69,13 +66,13 @@ public class InputTypeUtil {
}
if (convolutionMode == ConvolutionMode.Same) {
int hOut = stride[0] * hIn;
int wOut = stride[1] * wIn;
long hOut = stride[0] * hIn;
long wOut = stride[1] * wIn;
return InputType.convolutional(hOut, wOut, outputDepth);
}
int hOut = sH * (hIn - 1) + kH - 2 * padH;
int wOut = sW * (wIn - 1) + kW - 2 * padW;
long hOut = sH * (hIn - 1) + kH - 2 * padH;
long wOut = sW * (wIn - 1) + kW - 2 * padW;
return InputType.convolutional(hOut, wOut, outputDepth);
}
@ -91,10 +88,9 @@ public class InputTypeUtil {
InputType.InputTypeConvolutional3D i = (InputType.InputTypeConvolutional3D) inputType;
// FIXME: int cast
val inDepth = (int) i.getDepth();
val inHeight = (int) i.getHeight();
val inWidth = (int) i.getWidth();
long inDepth = i.getDepth();
long inHeight = i.getHeight();
long inWidth = i.getWidth();
int padD = (padding == null ? 0 : padding[0]);
int padH = (padding == null ? 0 : padding[1]);
@ -211,9 +207,9 @@ public class InputTypeUtil {
return InputType.convolutional3D(outD, outH, outW, outputChannels);
}
int dOut = (inDepth - kD + 2 * padD) / sD + 1;
int hOut = (inHeight - kH + 2 * padH) / sH + 1;
int wOut = (inWidth - kW + 2 * padW) / sW + 1;
long dOut = (inDepth - kD + 2 * padD) / sD + 1;
long hOut = (inHeight - kH + 2 * padH) / sH + 1;
long wOut = (inWidth - kW + 2 * padW) / sW + 1;
return InputType.convolutional3D(dOut, hOut, wOut, outputChannels);
}
@ -296,9 +292,8 @@ public class InputTypeUtil {
InputType.InputTypeConvolutional i = (InputType.InputTypeConvolutional) inputType;
// FIXME: int cast
val inHeight = (int) i.getHeight();
val inWidth = (int) i.getWidth();
long inHeight = i.getHeight();
long inWidth = i.getWidth();
int padH = (padding == null ? 0 : padding[0]); //May be null for ConvolutionMode.Same
int padW = (padding == null ? 0 : padding[1]);
int kH = kernelSize[0];
@ -379,8 +374,8 @@ public class InputTypeUtil {
return InputType.convolutional(outH, outW, outputDepth);
}
int hOut = (inHeight - kH + 2 * padH) / sH + 1;
int wOut = (inWidth - kW + 2 * padW) / sW + 1;
long hOut = (inHeight - kH + 2 * padH) / sH + 1;
long wOut = (inWidth - kW + 2 * padW) / sW + 1;
return InputType.convolutional(hOut, wOut, outputDepth);
}

4
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/conf/layers/RnnLossLayer.java
View File

@ -142,13 +142,13 @@ public class RnnLossLayer extends FeedForwardLayer {
}
@Override
public void setNIn(int nIn){
public void setNIn(long nIn){
throw new UnsupportedOperationException(
"This layer has no parameters, thus nIn will always equal nOut.");
}
@Override
public void setNOut(int nOut){
public void setNOut(long nOut){
throw new UnsupportedOperationException(
"This layer has no parameters, thus nIn will always equal nOut.");
}

4
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/conf/layers/Subsampling1DLayer.java
View File

@ -82,12 +82,12 @@ public class Subsampling1DLayer extends SubsamplingLayer {
}
InputType.InputTypeRecurrent r = (InputType.InputTypeRecurrent) inputType;
long inputTsLength = r.getTimeSeriesLength();
int outLength;
long outLength;
if (inputTsLength < 0) {
//Probably: user did InputType.recurrent(x) without specifying sequence length
outLength = -1;
} else {
outLength = Convolution1DUtils.getOutputSize((int) inputTsLength, kernelSize[0], stride[0], padding[0],
outLength = Convolution1DUtils.getOutputSize(inputTsLength, kernelSize[0], stride[0], padding[0],
convolutionMode, dilation[0]);
}
return InputType.recurrent(r.getSize(), outLength);

7
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/conf/layers/Subsampling3DLayer.java
View File

@ -32,6 +32,7 @@ import org.deeplearning4j.util.ValidationUtils;
import org.nd4j.base.Preconditions;
import org.nd4j.linalg.api.buffer.DataType;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.learning.regularization.Regularization;
import java.util.Collection;
@ -138,9 +139,11 @@ public class Subsampling3DLayer extends NoParamLayer {
+ "\"): Expected CNN input, got " + inputType);
}
// FIXME: int cast
long inChannels = ((InputType.InputTypeConvolutional3D) inputType).getChannels();
if (inChannels > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
return InputTypeUtil.getOutputTypeCnn3DLayers(inputType, kernelSize, stride, padding, new int[] {1, 1, 1}, // no dilation
convolutionMode, (int) ((InputType.InputTypeConvolutional3D) inputType).getChannels(),
convolutionMode, (int) inChannels,
layerIndex, getLayerName(), Subsampling3DLayer.class);
}

9
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/conf/layers/Upsampling3D.java
View File

@ -83,11 +83,10 @@ public class Upsampling3D extends BaseUpsamplingLayer {
}
InputType.InputTypeConvolutional3D i = (InputType.InputTypeConvolutional3D) inputType;
// FIXME: int cast
int inHeight = (int) i.getHeight();
int inWidth = (int) i.getWidth();
int inDepth = (int) i.getDepth();
int inChannels = (int) i.getChannels();
long inHeight = (int) i.getHeight();
long inWidth = (int) i.getWidth();
long inDepth = (int) i.getDepth();
long inChannels = (int) i.getChannels();
return InputType.convolutional3D(size[0] * inDepth, size[1] * inHeight, size[2] * inWidth, inChannels);
}

2
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/conf/ocnn/OCNNOutputLayer.java
View File

@ -259,7 +259,7 @@ public class OCNNOutputLayer extends BaseOutputLayer {
}
@Override
public void setNOut(int nOut){
public void setNOut(long nOut){
throw new UnsupportedOperationException(
"Unable to specify number of outputs with ocnn. Outputs are fixed to 1.");
}

22
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/graph/ComputationGraph.java
View File

@ -79,6 +79,7 @@ import org.nd4j.linalg.dataset.api.DataSetUtil;
import org.nd4j.linalg.dataset.api.MultiDataSet;
import org.nd4j.linalg.dataset.api.iterator.DataSetIterator;
import org.nd4j.linalg.dataset.api.iterator.MultiDataSetIterator;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.heartbeat.Heartbeat;
import org.nd4j.linalg.heartbeat.reports.Environment;
@ -3329,7 +3330,6 @@ public class ComputationGraph implements Serializable, Model, NeuralNetwork {
//In 99+% of cases, the input and labels dimension 0 size should be identical
//The only real exceptions: space to batch, and batch to space layers
//In those cases, we should base it on the labels size, as this impacts gradient calculation
// FIXME: int cast
return labels == null || labels[0] == null ? (int) inputs[0].size(0) : (int)labels[0].size(0);
}
@ -3653,7 +3653,8 @@ public class ComputationGraph implements Serializable, Model, NeuralNetwork {
if (endTimeIdx > timeSeriesLength)
endTimeIdx = timeSeriesLength;
// FIXME: int cast
if (startTimeIdx > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
List<INDArray[]> list = getSubsetsForTbptt((int) startTimeIdx, endTimeIdx, inputs, labels, featureMasks, labelMasks);
setInputs(list.get(0));
@ -3799,7 +3800,8 @@ public class ComputationGraph implements Serializable, Model, NeuralNetwork {
}
}
// FIXME: int cast
if (minibatchSize > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
Pair<INDArray, MaskState> outPair =
current.feedForwardMaskArrays(inputMasks, maskState, (int)minibatchSize);
map.put(topologicalOrder[i], outPair);
@ -4664,7 +4666,7 @@ public class ComputationGraph implements Serializable, Model, NeuralNetwork {
* @param layer Index of the layer to get the size of. Must be in range 0 to nLayers-1 inclusive
* @return Size of the layer
*/
public int layerSize(int layer) {
public long layerSize(int layer) {
if (layer < 0 || layer > layers.length) {
throw new IllegalArgumentException("Invalid layer index: " + layer + ". Layer index must be between 0 and "
+ (layers.length - 1) + " inclusive");
@ -4683,7 +4685,7 @@ public class ComputationGraph implements Serializable, Model, NeuralNetwork {
* @param layer Index of the layer to get the size of. Must be in range 0 to nLayers-1 inclusive
* @return Size of the layer
*/
public int layerInputSize(int layer) {
public long layerInputSize(int layer) {
if (layer < 0 || layer > layers.length) {
throw new IllegalArgumentException("Invalid layer index: " + layer + ". Layer index must be between 0 and "
+ (layers.length - 1) + " inclusive");
@ -4701,7 +4703,7 @@ public class ComputationGraph implements Serializable, Model, NeuralNetwork {
* @param layerName Name of the layer to get the size of
* @return Size of the layer
*/
public int layerSize(String layerName) {
public long layerSize(String layerName) {
Layer l = getLayer(layerName);
if(l == null){
throw new IllegalArgumentException("No layer with name \"" + layerName + "\" exists");
@ -4712,8 +4714,7 @@ public class ComputationGraph implements Serializable, Model, NeuralNetwork {
}
FeedForwardLayer ffl = (FeedForwardLayer) conf;
// FIXME: int cast
return (int) ffl.getNOut();
return ffl.getNOut();
}
/**
@ -4727,7 +4728,7 @@ public class ComputationGraph implements Serializable, Model, NeuralNetwork {
* @param layerName Name of the layer to get the size of
* @return Size of the layer
*/
public int layerInputSize(String layerName) {
public long layerInputSize(String layerName) {
Layer l = getLayer(layerName);
if(l == null){
throw new IllegalArgumentException("No layer with name \"" + layerName + "\" exists");
@ -4738,8 +4739,7 @@ public class ComputationGraph implements Serializable, Model, NeuralNetwork {
}
FeedForwardLayer ffl = (FeedForwardLayer) conf;
// FIXME: int cast
return (int) ffl.getNIn();
return ffl.getNIn();
}
/**

19
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/graph/vertex/impl/UnstackVertex.java
View File

@ -43,10 +43,10 @@ import java.util.Arrays;
* @author Justin Long (crockpotveggies)
*/
public class UnstackVertex extends BaseGraphVertex {
private int from;
private long from;
private int stackSize;
private long forwardShape[];
private int step;
private long step;
public UnstackVertex(ComputationGraph graph, String name, int vertexIndex, int from, int stackSize, DataType dataType) {
this(graph, name, vertexIndex, null, null, from, stackSize, dataType);
@ -77,10 +77,9 @@ public class UnstackVertex extends BaseGraphVertex {
// once we know the inputs, save the shape and interval size for doBackward
this.forwardShape = Arrays.copyOf(inputs[0].shape(), inputs[0].rank());
// FIXME: int cast
this.step = (int) inputs[0].size(0) / stackSize;
int start = from * step;
int end = (from + 1) * step;
this.step = inputs[0].size(0) / stackSize;
long start = from * step;
long end = (from + 1) * step;
INDArray ret;
switch (inputs[0].rank()) { //TODO remove the dups here if/when possible (gradient checks must pass)
@ -108,8 +107,8 @@ public class UnstackVertex extends BaseGraphVertex {
throw new IllegalStateException("Cannot do backward pass: error not set");
INDArray out = workspaceMgr.create(ArrayType.ACTIVATION_GRAD, inputs[0].dataType(), forwardShape);
int start = from * step;
int end = (from + 1) * step;
long start = from * step;
long end = (from + 1) * step;
switch (forwardShape.length) {
case 2:
@ -154,8 +153,8 @@ public class UnstackVertex extends BaseGraphVertex {
}
//Mask arrays are either 1d (column vector) or 2d...
int start = from * minibatchSize;
int end = (from + 1) * minibatchSize;
long start = from * minibatchSize;
long end = (from + 1) * minibatchSize;
INDArray outMask = maskArrays[0].get(NDArrayIndex.interval(start, end), NDArrayIndex.all());
return new Pair<>(outMask, currentMaskState);
}

6
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/graph/vertex/impl/rnn/LastTimeStepVertex.java
View File

@ -87,9 +87,8 @@ public class LastTimeStepVertex extends BaseGraphVertex {
INDArray out;
if (mask == null) {
// FIXME: int cast
//No mask array -> extract same (last) column for all
int lastTS = (int) inputs[0].size(2) - 1;
long lastTS = inputs[0].size(2) - 1;
out = inputs[0].get(NDArrayIndex.all(), NDArrayIndex.all(), NDArrayIndex.point(lastTS));
out = workspaceMgr.dup(ArrayType.ACTIVATIONS, out);
fwdPassTimeSteps = null; //Null -> last time step for all examples
@ -99,8 +98,7 @@ public class LastTimeStepVertex extends BaseGraphVertex {
//Want the index of the last non-zero entry in the mask array.
//Check a little here by using mulRowVector([0,1,2,3,...]) and argmax
// FIXME: int cast
int maxTsLength = (int) fwdPassShape[2];
long maxTsLength = fwdPassShape[2];
INDArray row = Nd4j.linspace(0, maxTsLength - 1, maxTsLength, mask.dataType());
INDArray temp = mask.mulRowVector(row);
INDArray lastElementIdx = Nd4j.argMax(temp, 1);

1
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/AbstractLayer.java
View File

@ -346,7 +346,6 @@ public abstract class AbstractLayer<LayerConfT extends org.deeplearning4j.nn.con
@Override
public int getInputMiniBatchSize() {
// FIXME: int cast
return (int) input.size(0);
}

1
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/BaseOutputLayer.java
View File

@ -229,7 +229,6 @@ public abstract class BaseOutputLayer<LayerConfT extends org.deeplearning4j.nn.c
*/
@Override
public int numLabels() {
// FIXME: int cast
return (int) labels.size(1);
}

1
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/LossLayer.java
View File

@ -236,7 +236,6 @@ public class LossLayer extends BaseLayer<org.deeplearning4j.nn.conf.layers.LossL
*/
@Override
public int numLabels() {
// FIXME: int cast
return (int) labels.size(1);
}

49
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/convolution/Cnn3DLossLayer.java
View File

@ -86,19 +86,18 @@ public class Cnn3DLossLayer extends BaseLayer<org.deeplearning4j.nn.conf.layers.
INDArray delta2d = lossFunction.computeGradient(labels2d, input2d.dup(input2d.ordering()), layerConf().getActivationFn(), maskReshaped);
delta2d = workspaceMgr.leverageTo(ArrayType.ACTIVATION_GRAD, delta2d);
// FIXME: int cast
int n = (int)input.size(0);
int d, h, w, c;
long n = input.size(0);
long d, h, w, c;
if(layerConf().getDataFormat() == Convolution3D.DataFormat.NDHWC){
d = (int)input.size(1);
h = (int)input.size(2);
w = (int)input.size(3);
c = (int)input.size(4);
d = input.size(1);
h = input.size(2);
w = input.size(3);
c = input.size(4);
} else {
d = (int)input.size(2);
h = (int)input.size(3);
w = (int)input.size(4);
c = (int)input.size(1);
d = input.size(2);
h = input.size(3);
w = input.size(4);
c = input.size(1);
}
INDArray delta5d = ConvolutionUtils.reshape2dTo5d(layerConf().getDataFormat(), delta2d, n, d, h, w, c, workspaceMgr, ArrayType.ACTIVATION_GRAD);
@ -130,7 +129,6 @@ public class Cnn3DLossLayer extends BaseLayer<org.deeplearning4j.nn.conf.layers.
@Override
public int numLabels() {
// FIXME: int cast
return (int) labels.size(1);
}
@ -180,10 +178,8 @@ public class Cnn3DLossLayer extends BaseLayer<org.deeplearning4j.nn.conf.layers.
INDArray input2d = ConvolutionUtils.reshape5dTo2d(layerConf().getDataFormat(), in, workspaceMgr, ArrayType.ACTIVATIONS);
INDArray out2d = layerConf().getActivationFn().getActivation(input2d, training);
// FIXME: int cast
int n = (int)input.size(0);
int d, h, w, c;
long n = input.size(0);
long d, h, w, c;
if(layerConf().getDataFormat() == Convolution3D.DataFormat.NDHWC){
d = (int)input.size(1);
h = (int)input.size(2);
@ -262,19 +258,18 @@ public class Cnn3DLossLayer extends BaseLayer<org.deeplearning4j.nn.conf.layers.
val newShape = input.shape().clone();
newShape[1] = 1;
// FIXME
int n = (int)input.size(0);
int d, h, w, c;
long n = input.size(0);
long d, h, w, c;
if(layerConf().getDataFormat() == Convolution3D.DataFormat.NDHWC){
d = (int)input.size(1);
h = (int)input.size(2);
w = (int)input.size(3);
c = (int)input.size(4);
d = input.size(1);
h = input.size(2);
w = input.size(3);
c = input.size(4);
} else {
d = (int)input.size(2);
h = (int)input.size(3);
w = (int)input.size(4);
c = (int)input.size(1);
d = input.size(2);
h = input.size(3);
w = input.size(4);
c = input.size(1);
}
INDArray scoreArrayTs = ConvolutionUtils.reshape2dTo5d(layerConf().getDataFormat(), scoreArray, n, d, h, w, c, workspaceMgr, ArrayType.FF_WORKING_MEM);
INDArray summedScores = scoreArrayTs.sum(1,2,3,4);

10
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/convolution/CnnLossLayer.java
View File

@ -88,8 +88,7 @@ public class CnnLossLayer extends BaseLayer<org.deeplearning4j.nn.conf.layers.Cn
INDArray delta2d = lossFunction.computeGradient(labels2d, input2d.dup(input2d.ordering()), layerConf().getActivationFn(), maskReshaped);
delta2d = workspaceMgr.leverageTo(ArrayType.ACTIVATION_GRAD, delta2d);
// FIXME: int cast
INDArray delta4d = ConvolutionUtils.reshape2dTo4d(delta2d, ArrayUtil.toInts(input.shape()), workspaceMgr, ArrayType.ACTIVATION_GRAD);
INDArray delta4d = ConvolutionUtils.reshape2dTo4d(delta2d, input.shape(), workspaceMgr, ArrayType.ACTIVATION_GRAD);
// grab the empty gradient
Gradient gradient = new DefaultGradient();
@ -119,7 +118,6 @@ public class CnnLossLayer extends BaseLayer<org.deeplearning4j.nn.conf.layers.Cn
@Override
public int numLabels() {
// FIXME: int cast
return (int) labels.size(1);
}
@ -169,8 +167,7 @@ public class CnnLossLayer extends BaseLayer<org.deeplearning4j.nn.conf.layers.Cn
INDArray input2d = ConvolutionUtils.reshape4dTo2d(in, workspaceMgr, ArrayType.ACTIVATIONS);
INDArray out2d = layerConf().getActivationFn().getActivation(input2d, training);
// FIXME: int cast
return ConvolutionUtils.reshape2dTo4d(out2d, ArrayUtil.toInts(input.shape()), workspaceMgr, ArrayType.ACTIVATIONS);
return ConvolutionUtils.reshape2dTo4d(out2d, input.shape(), workspaceMgr, ArrayType.ACTIVATIONS);
}
@Override
@ -236,8 +233,7 @@ public class CnnLossLayer extends BaseLayer<org.deeplearning4j.nn.conf.layers.Cn
val newShape = input.shape().clone();
newShape[1] = 1;
// FIXME
INDArray scoreArrayTs = ConvolutionUtils.reshape2dTo4d(scoreArray, ArrayUtil.toInts(newShape), workspaceMgr, ArrayType.FF_WORKING_MEM);
INDArray scoreArrayTs = ConvolutionUtils.reshape2dTo4d(scoreArray, newShape, workspaceMgr, ArrayType.FF_WORKING_MEM);
INDArray summedScores = scoreArrayTs.sum(1,2,3).reshape(scoreArrayTs.size(0), 1);
if (fullNetRegTerm != 0.0) {

6
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/convolution/Convolution3DLayer.java
View File

@ -71,8 +71,7 @@ public class Convolution3DLayer extends ConvolutionLayer {
boolean isNCDHW = layerConfig.getDataFormat() == Convolution3D.DataFormat.NCDHW;
// FIXME: int cast
int miniBatch = (int) input.size(0);
long miniBatch = input.size(0);
int inD = (int) (isNCDHW ? input.size(2) : input.size(1));
int inH = (int) (isNCDHW ? input.size(3) : input.size(2));
int inW = (int) (isNCDHW ? input.size(4) : input.size(3));
@ -189,8 +188,7 @@ public class Convolution3DLayer extends ConvolutionLayer {
+ " " + layerId());
}
// FIXME: int cast
int miniBatch = (int) input.size(0);
long miniBatch = input.size(0);
int inputChannels = (int) (isNCDHW ? input.size(1) : input.size(4));
int inD =(int) (isNCDHW ? input.size(2) : input.size(1));
int inH = (int) (isNCDHW ? input.size(3) : input.size(2));

38
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/convolution/ConvolutionLayer.java
View File

@ -35,6 +35,7 @@ import org.nd4j.linalg.api.memory.MemoryWorkspace;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.api.shape.Shape;
import org.nd4j.linalg.convolution.Convolution;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.exception.ND4JOpProfilerException;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.primitives.Pair;
@ -113,13 +114,12 @@ public class ConvolutionLayer extends BaseLayer<org.deeplearning4j.nn.conf.layer
if(epsilon.dataType() != dataType)
epsilon = epsilon.castTo(dataType);
// FIXME: int cast
int miniBatch = (int) input.size(0);
long miniBatch = input.size(0);
int inH = (int) input.size(2);
int inW = (int) input.size(3);
int outDepth = (int) weights.size(0);
int inDepth = (int) weights.size(1);
long outDepth = weights.size(0);
long inDepth = weights.size(1);
int kH = (int) weights.size(2);
int kW = (int) weights.size(3);
@ -143,7 +143,7 @@ public class ConvolutionLayer extends BaseLayer<org.deeplearning4j.nn.conf.layer
INDArray biasGradView = gradientViews.get(ConvolutionParamInitializer.BIAS_KEY);
INDArray weightGradView = gradientViews.get(ConvolutionParamInitializer.WEIGHT_KEY); //4d, c order. Shape: [outDepth,inDepth,kH,kW]
INDArray weightGradView2df = Shape
.newShapeNoCopy(weightGradView, new int[] {outDepth, inDepth * kH * kW}, false).transpose();
.newShapeNoCopy(weightGradView, new long[]{outDepth, inDepth * kH * kW}, false).transpose();
@ -204,7 +204,7 @@ public class ConvolutionLayer extends BaseLayer<org.deeplearning4j.nn.conf.layer
//Note: due to the permute in preOut, and the fact that we essentially do a preOut.muli(epsilon), this reshape
// should be zero-copy; only possible exception being sometimes with the "identity" activation case
INDArray delta2d = delta.reshape('c', new int[] {outDepth, miniBatch * outH * outW}); //Shape.newShapeNoCopy(delta,new int[]{outDepth,miniBatch*outH*outW},false);
INDArray delta2d = delta.reshape('c', new long[] {outDepth, miniBatch * outH * outW}); //Shape.newShapeNoCopy(delta,new int[]{outDepth,miniBatch*outH*outW},false);
//Do im2col, but with order [miniB,outH,outW,depthIn,kH,kW]; but need to input [miniBatch,channels,kH,kW,outH,outW] given the current im2col implementation
//To get this: create an array of the order we want, permute it to the order required by im2col implementation, and then do im2col on that
@ -231,7 +231,7 @@ public class ConvolutionLayer extends BaseLayer<org.deeplearning4j.nn.conf.layer
//Calculate epsilons for layer below, in 2d format (note: this is in 'image patch' format before col2im reduction)
//Note: cc -> f mmul here, then reshape to 6d in f order
INDArray epsNext2d = w2d.mmul(delta2d); //TODO can we reuse im2col array instead of allocating new result array?
INDArray eps6d = Shape.newShapeNoCopy(epsNext2d, new int[] {kW, kH, inDepth, outW, outH, miniBatch}, true);
INDArray eps6d = Shape.newShapeNoCopy(epsNext2d, new long[] {kW, kH, inDepth, outW, outH, miniBatch}, true);
//Calculate epsilonNext by doing im2col reduction.
//Current col2im implementation expects input with order: [miniBatch,channels,kH,kW,outH,outW]
@ -282,7 +282,7 @@ public class ConvolutionLayer extends BaseLayer<org.deeplearning4j.nn.conf.layer
}
}
protected void validateInputDepth(int inDepth) {
protected void validateInputDepth(long inDepth) {
if (input.size(1) != inDepth) {
String layerName = conf.getLayer().getLayerName();
if (layerName == null)
@ -313,14 +313,13 @@ public class ConvolutionLayer extends BaseLayer<org.deeplearning4j.nn.conf.layer
INDArray input = this.input.castTo(dataType);
// FIXME: int cast
int miniBatch = (int) input.size(0);
int outDepth = (int) weights.size(0);
int inDepth = (int) weights.size(1);
long miniBatch = input.size(0);
long outDepth = weights.size(0);
long inDepth = weights.size(1);
validateInputDepth(inDepth);
int kH = (int) weights.size(2);
int kW = (int) weights.size(3);
long kH = weights.size(2);
long kW = weights.size(3);
int[] dilation = layerConf().getDilation();
int[] kernel = layerConf().getKernelSize();
@ -331,7 +330,8 @@ public class ConvolutionLayer extends BaseLayer<org.deeplearning4j.nn.conf.layer
if (convolutionMode == ConvolutionMode.Same) {
outSize = ConvolutionUtils.getOutputSize(input, kernel, strides, null, convolutionMode, dilation); //Also performs validation
// FIXME: int cast
if (input.size(2) > Integer.MAX_VALUE || input.size(3) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
pad = ConvolutionUtils.getSameModeTopLeftPadding(outSize, new int[] {(int) input.size(2), (int) input.size(3)}, kernel,
strides, dilation );
} else {
@ -397,10 +397,12 @@ public class ConvolutionLayer extends BaseLayer<org.deeplearning4j.nn.conf.layer
INDArray col = Nd4j.createUninitialized(weights.dataType(), new long[] {miniBatch, outH, outW, inDepth, kH, kW}, 'c');
INDArray col2 = col.permute(0, 3, 4, 5, 1, 2);
INDArray im2ColIn = input.castTo(col2.dataType()); //No op if already (for example) float
Convolution.im2col(im2ColIn, kH, kW, strides[0], strides[1], pad[0], pad[1], dilation[0], dilation[1],
if (kH > Integer.MAX_VALUE || kW > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
Convolution.im2col(im2ColIn, (int)kH, (int)kW, strides[0], strides[1], pad[0], pad[1], dilation[0], dilation[1],
convolutionMode == ConvolutionMode.Same, col2);
INDArray im2col2d = Shape.newShapeNoCopy(col, new int[] {miniBatch * outH * outW, inDepth * kH * kW}, false);
INDArray im2col2d = Shape.newShapeNoCopy(col, new long[] {miniBatch * outH * outW, inDepth * kH * kW}, false);
//Current order of weights: [depthOut,depthIn,kH,kW], c order
//Permute to give [kW,kH,depthIn,depthOut], f order
@ -418,7 +420,7 @@ public class ConvolutionLayer extends BaseLayer<org.deeplearning4j.nn.conf.layer
}
//Now, reshape to [outW,outH,miniBatch,outDepth], and permute to have correct output order: [miniBath,outDepth,outH,outW];
z = Shape.newShapeNoCopy(z, new int[] {outW, outH, miniBatch, outDepth}, true);
z = Shape.newShapeNoCopy(z, new long[] {outW, outH, miniBatch, outDepth}, true);
z = z.permute(2, 3, 1, 0);
if (training && cacheMode != CacheMode.NONE && workspaceMgr.hasConfiguration(ArrayType.FF_CACHE) && workspaceMgr.isWorkspaceOpen(ArrayType.FF_CACHE)) {

10
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/convolution/Deconvolution2DLayer.java
View File

@ -171,9 +171,8 @@ public class Deconvolution2DLayer extends ConvolutionLayer {
+ " " + layerId());
}
// FIXME: int cast
int inDepth = (int) weights.size(0);
int outDepth = (int) weights.size(1);
long inDepth = weights.size(0);
long outDepth = weights.size(1);
if (input.size(1) != inDepth && input.size(3) == inDepth) {
//TODO AB 2019/10/25 this is an ugly "pseudo-NHWC support" hack that needs to be removed ASAD
@ -199,7 +198,6 @@ public class Deconvolution2DLayer extends ConvolutionLayer {
int[] pad;
int[] outSize;
if (convolutionMode == ConvolutionMode.Same) {
// FIXME: int cast
outSize = ConvolutionUtils.getDeconvolutionOutputSize(input, kernel, strides, null, convolutionMode, dilation); //Also performs validation
pad = ConvolutionUtils.getSameModeTopLeftPadding(outSize, new int[] {(int) input.size(2), (int) input.size(3)}, kernel,
strides, dilation );
@ -208,8 +206,8 @@ public class Deconvolution2DLayer extends ConvolutionLayer {
outSize = ConvolutionUtils.getDeconvolutionOutputSize(input, kernel, strides, pad, convolutionMode, dilation); //Also performs validation
}
int outH = outSize[0];
int outW = outSize[1];
long outH = outSize[0];
long outW = outSize[1];
val miniBatch = input.size(0);

21
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/convolution/DepthwiseConvolution2DLayer.java
View File

@ -32,6 +32,7 @@ import org.nd4j.linalg.api.buffer.DataType;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.api.ops.CustomOp;
import org.nd4j.linalg.api.ops.DynamicCustomOp;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.primitives.Pair;
@ -75,12 +76,11 @@ public class DepthwiseConvolution2DLayer extends ConvolutionLayer {
INDArray input = this.input.castTo(dataType); //No-op if correct type
// FIXME: int cast
int miniBatch = (int) input.size(0);
long miniBatch = input.size(0);
int inH = (int)input.size(2);
int inW = (int)input.size(3);
int inDepth = (int) depthWiseWeights.size(2);
long inDepth = depthWiseWeights.size(2);
int kH = (int) depthWiseWeights.size(0);
int kW = (int) depthWiseWeights.size(1);
@ -169,10 +169,9 @@ public class DepthwiseConvolution2DLayer extends ConvolutionLayer {
INDArray input = this.input.castTo(dataType); //no-op if correct dtype
// FIXME: int cast
int inDepth = (int) depthWiseWeights.size(2);
int depthMultiplier = (int) depthWiseWeights.size(3);
int outDepth = depthMultiplier * inDepth;
long inDepth = depthWiseWeights.size(2);
long depthMultiplier = depthWiseWeights.size(3);
long outDepth = depthMultiplier * inDepth;
if (input.size(1) != inDepth) {
String layerName = conf.getLayer().getLayerName();
@ -197,7 +196,9 @@ public class DepthwiseConvolution2DLayer extends ConvolutionLayer {
if (convolutionMode == ConvolutionMode.Same) {
outSize = ConvolutionUtils.getOutputSize(input, kernel, strides, null, convolutionMode, dilation);
// FIXME: int cast
if (input.size(2) > Integer.MAX_VALUE || input.size(3) > Integer.MAX_VALUE) {
throw new ND4JArraySizeException();
}
pad = ConvolutionUtils.getSameModeTopLeftPadding(
outSize, new int[]{(int) input.size(2), (int) input.size(3)}, kernel, strides, dilation);
} else {
@ -205,8 +206,8 @@ public class DepthwiseConvolution2DLayer extends ConvolutionLayer {
outSize = ConvolutionUtils.getOutputSize(input, kernel, strides, pad, convolutionMode, dilation);
}
int outH = outSize[0];
int outW = outSize[1];
long outH = outSize[0];
long outW = outSize[1];
val miniBatch = input.size(0);
INDArray output = workspaceMgr.create(

13
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/convolution/SeparableConvolution2DLayer.java
View File

@ -33,6 +33,7 @@ import org.nd4j.linalg.api.memory.MemoryWorkspace;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.api.ops.CustomOp;
import org.nd4j.linalg.api.ops.DynamicCustomOp;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.primitives.Pair;
import org.deeplearning4j.nn.workspace.LayerWorkspaceMgr;
@ -90,8 +91,7 @@ public class SeparableConvolution2DLayer extends ConvolutionLayer {
INDArray input = this.input.castTo(dataType);
// FIXME: int cast
int miniBatch = (int) input.size(0);
long miniBatch = input.size(0);
int inH = (int)input.size(2);
int inW = (int)input.size(3);
@ -194,9 +194,8 @@ public class SeparableConvolution2DLayer extends ConvolutionLayer {
+ " " + layerId());
}
// FIXME: int cast
int inDepth = (int) depthWiseWeights.size(1);
int outDepth = (int) pointWiseWeights.size(0);
long inDepth = depthWiseWeights.size(1);
long outDepth = pointWiseWeights.size(0);
if (input.size(1) != inDepth) {
String layerName = conf.getLayer().getLayerName();
@ -220,7 +219,9 @@ public class SeparableConvolution2DLayer extends ConvolutionLayer {
if (convolutionMode == ConvolutionMode.Same) {
outSize = ConvolutionUtils.getOutputSize(input, kernel, strides, null, convolutionMode, dilation); //Also performs validation
// FIXME: int cast
if (input.size(2) > Integer.MAX_VALUE || input.size(3) > Integer.MAX_VALUE) {
throw new ND4JArraySizeException();
}
pad = ConvolutionUtils.getSameModeTopLeftPadding(outSize, new int[] {(int) input.size(2), (int) input.size(3)}, kernel,
strides, dilation );
} else {

24
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/convolution/SpaceToDepth.java
View File

@ -75,11 +75,10 @@ public class SpaceToDepth extends AbstractLayer<org.deeplearning4j.nn.conf.layer
public Pair<Gradient, INDArray> backpropGradient(INDArray epsilon, LayerWorkspaceMgr workspaceMgr) {
assertInputSet(true);
// FIXME: int cast
int miniBatch = (int) input.size(0);
int inDepth = (int) input.size(1);
int inH = (int) input.size(2);
int inW = (int) input.size(3);
long miniBatch = input.size(0);
long inDepth = input.size(1);
long inH = input.size(2);
long inW = input.size(3);
INDArray input = this.input.castTo(dataType); //No-op if already correct type
@ -122,17 +121,16 @@ public class SpaceToDepth extends AbstractLayer<org.deeplearning4j.nn.conf.layer
return preOutput;
}
// FIXME: int cast
int miniBatch = (int) input.size(0);
int depth = (int) input.size(1);
int inH = (int) input.size(2);
int inW = (int) input.size(3);
long miniBatch = input.size(0);
long depth = input.size(1);
long inH = input.size(2);
long inW = input.size(3);
int blockSize = getBlockSize();
int outH = inH / blockSize;
int outW = inW / blockSize;
int outDepth = depth * blockSize * blockSize;
long outH = inH / blockSize;
long outW = inW / blockSize;
long outDepth = depth * blockSize * blockSize;
INDArray out = workspaceMgr.create(ArrayType.ACTIVATIONS, input.dataType(), new long[]{1, miniBatch * outDepth * outH * outW}, 'c');
INDArray reshapedOut;

16
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/convolution/subsampling/Subsampling3DLayer.java
View File

@ -71,9 +71,8 @@ public class Subsampling3DLayer extends AbstractLayer<org.deeplearning4j.nn.conf
boolean isNCDHW = layerConf().getDataFormat() == Convolution3D.DataFormat.NCDHW;
// FIXME: int cast
int miniBatch = (int) input.size(0);
int inChannels = (int) (isNCDHW ? input.size(1) : input.size(4));
long miniBatch = input.size(0);
long inChannels = isNCDHW ? input.size(1) : input.size(4);
int inD = (int) (isNCDHW ? input.size(2) : input.size(1));
int inH = (int) (isNCDHW ? input.size(3) : input.size(2));
int inW = (int) (isNCDHW ? input.size(4) : input.size(3));
@ -148,9 +147,8 @@ public class Subsampling3DLayer extends AbstractLayer<org.deeplearning4j.nn.conf
}
}
// FIXME: int cast
int miniBatch = (int) input.size(0);
int inChannels = (int) (isNCDHW ? input.size(1) : input.size(4));
long miniBatch = input.size(0);
long inChannels = isNCDHW ? input.size(1) : input.size(4);
int inD = (int) (isNCDHW ? input.size(2) : input.size(1));
int inH = (int) (isNCDHW ? input.size(3) : input.size(2));
int inW = (int) (isNCDHW ? input.size(4) : input.size(3));
@ -170,9 +168,9 @@ public class Subsampling3DLayer extends AbstractLayer<org.deeplearning4j.nn.conf
outSize = Convolution3DUtils.get3DOutputSize(
input, kernel, strides, pad, convolutionMode, dilation, isNCDHW);
}
int outD = outSize[0];
int outH = outSize[1];
int outW = outSize[2];
long outD = outSize[0];
long outH = outSize[1];
long outW = outSize[2];
String opName = layerConf().getPoolingType() == PoolingType.MAX ? "maxpool3dnew" : "avgpool3dnew";

18
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/convolution/subsampling/SubsamplingLayer.java
View File

@ -108,9 +108,6 @@ public class SubsamplingLayer extends AbstractLayer<org.deeplearning4j.nn.conf.l
if(epsilon.dataType() != dataType)
epsilon = epsilon.castTo(dataType);
// FIXME: int cast
int miniBatch = (int) input.size(0);
int inDepth = (int) input.size(1);
int inH = (int)input.size(2);
int inW = (int)input.size(3);
@ -158,9 +155,6 @@ public class SubsamplingLayer extends AbstractLayer<org.deeplearning4j.nn.conf.l
//subsampling doesn't have weights and thus gradients are not calculated for this layer
//only scale and reshape epsilon
// FIXME: int cast
int inputHeight = (int) input().size(-2);
int inputWidth = (int) input().size(-1);
Gradient retGradient = new DefaultGradient();
@ -231,9 +225,8 @@ public class SubsamplingLayer extends AbstractLayer<org.deeplearning4j.nn.conf.l
INDArray input = this.input.castTo(dataType);
// FIXME: int cast
int miniBatch = (int) input.size(0);
int inDepth = (int) input.size(1);
long miniBatch = input.size(0);
long inDepth = input.size(1);
int inH = (int)input.size(2);
int inW = (int)input.size(3);
@ -250,8 +243,8 @@ public class SubsamplingLayer extends AbstractLayer<org.deeplearning4j.nn.conf.l
pad = layerConf().getPadding();
outSize = ConvolutionUtils.getOutputSize(input, kernel, strides, pad, convolutionMode, dilation); //Also performs validation
}
int outH = outSize[0];
int outW = outSize[1];
long outH = outSize[0];
long outW = outSize[1];
if (helper != null && (helperCountFail == 0 || !layerConf().isCudnnAllowFallback())) {
@ -278,9 +271,6 @@ public class SubsamplingLayer extends AbstractLayer<org.deeplearning4j.nn.conf.l
}
}
INDArray output = workspaceMgr.createUninitialized(ArrayType.ACTIVATIONS, input.dataType(), new long[]{miniBatch, inDepth, outH, outW}, 'c');
DynamicCustomOp.DynamicCustomOpsBuilder b;
int extra = 0;

9
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/convolution/upsampling/Upsampling1D.java
View File

@ -65,11 +65,10 @@ public class Upsampling1D extends Upsampling2D {
INDArray originalInput = input;
input = input.castTo(dataType).reshape(input.size(0), input.size(1), input.size(2), 1);
// FIXME: int cast
int miniBatch = (int) input.size(0);
int inDepth = (int) input.size(1);
int inH = (int) input.size(2);
int inW = (int) input.size(3);
long miniBatch = input.size(0);
long inDepth = input.size(1);
long inH = input.size(2);
long inW = input.size(3);
INDArray outEpsilon = workspaceMgr.createUninitialized(ArrayType.ACTIVATION_GRAD, input.dataType(), miniBatch * inDepth * inH * inW);

22
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/convolution/upsampling/Upsampling2D.java
View File

@ -62,11 +62,10 @@ public class Upsampling2D extends AbstractLayer<org.deeplearning4j.nn.conf.layer
public Pair<Gradient, INDArray> backpropGradient(INDArray epsilon, LayerWorkspaceMgr workspaceMgr) {
assertInputSet(true);
// FIXME: int cast
int miniBatch = (int) input.size(0);
int inDepth = (int) input.size(1);
int inH = (int) input.size(2);
int inW = (int) input.size(3);
long miniBatch = (int) input.size(0);
long inDepth = (int) input.size(1);
long inH = (int) input.size(2);
long inW = (int) input.size(3);
INDArray reshapedEpsilon = workspaceMgr.createUninitialized(ArrayType.ACTIVATION_GRAD, epsilon.dataType(), new long[]{miniBatch, inDepth, inH, inW}, 'c');
@ -106,15 +105,14 @@ public class Upsampling2D extends AbstractLayer<org.deeplearning4j.nn.conf.layer
return preOutput;
}
// FIXME: int cast
int miniBatch = (int) input.size(0);
int inDepth = (int) input.size(1);
int inH = (int) input.size(2);
int inW = (int) input.size(3);
long miniBatch = (int) input.size(0);
long inDepth = (int) input.size(1);
long inH = (int) input.size(2);
long inW = (int) input.size(3);
int[] size = getSize();
int outH = inH * size[0];
int outW = inW * size[1];
int outH = (int)inH * size[0];
int outW = (int)inW * size[1];
INDArray reshapedOutput = workspaceMgr.createUninitialized(ArrayType.ACTIVATIONS, input.dataType(), new long[]{miniBatch, inDepth, outH, outW}, 'c');

26
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/convolution/upsampling/Upsampling3D.java
View File

@ -68,22 +68,21 @@ public class Upsampling3D extends AbstractLayer<org.deeplearning4j.nn.conf.layer
assertInputSet(true);
boolean ncdhw = layerConf().getDataFormat() == org.deeplearning4j.nn.conf.layers.Convolution3D.DataFormat.NCDHW;
// FIXME: int cast
// Assumes NCDHW order
int miniBatch = (int) input.size(0);
int inChannels, inD, inH, inW;
long miniBatch = input.size(0);
long inChannels, inD, inH, inW;
int[] intArgs;
if(ncdhw){
inChannels = (int) input.size(1);
inD = (int) input.size(2);
inH = (int) input.size(3);
inW = (int) input.size(4);
inChannels = input.size(1);
inD = input.size(2);
inH = input.size(3);
inW = input.size(4);
intArgs = new int[] {1}; // 1 is channels first
} else {
inD = (int) input.size(1);
inH = (int) input.size(2);
inW = (int) input.size(3);
inChannels = (int) input.size(4);
inD = input.size(1);
inH = input.size(2);
inW = input.size(3);
inChannels = input.size(4);
intArgs = new int[] {0}; // 0 is channels last
}
@ -134,9 +133,8 @@ public class Upsampling3D extends AbstractLayer<org.deeplearning4j.nn.conf.layer
}
boolean ncdhw = layerConf().getDataFormat() == org.deeplearning4j.nn.conf.layers.Convolution3D.DataFormat.NCDHW;
// FIXME: int cast
int miniBatch = (int) input.size(0);
int inChannels, inD, inH, inW;
long miniBatch = input.size(0);
long inChannels, inD, inH, inW;
int[] intArgs;
int[] size = getSize();
if(ncdhw){

7
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/feedforward/embedding/EmbeddingLayer.java
View File

@ -20,6 +20,7 @@ import lombok.extern.slf4j.Slf4j;
import lombok.val;
import org.nd4j.linalg.api.buffer.DataType;
import org.nd4j.linalg.api.ops.custom.ScatterUpdate;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.primitives.Pair;
import org.deeplearning4j.exception.DL4JInvalidInputException;
import org.deeplearning4j.nn.conf.NeuralNetConfiguration;
@ -64,8 +65,7 @@ public class EmbeddingLayer extends BaseLayer<org.deeplearning4j.nn.conf.layers.
INDArray weightGradients = gradientViews.get(DefaultParamInitializer.WEIGHT_KEY);
weightGradients.assign(0);
// FIXME: int cast
int[] indexes = new int[(int) input.length()];
long[] indexes = new long[(int) input.length()];
for (int i = 0; i < indexes.length; i++) {
indexes[i] = input.getInt(i, 0);
}
@ -99,7 +99,8 @@ public class EmbeddingLayer extends BaseLayer<org.deeplearning4j.nn.conf.layers.
val nIn = layerConf().getNIn();
// FIXME: int cast
if (input.length() > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
int[] indexes = new int[(int) input.length()];
for (int i = 0; i < indexes.length; i++) {
indexes[i] = input.getInt(i, 0);

4
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/mkldnn/MKLDNNBatchNormHelper.java
View File

@ -56,7 +56,7 @@ public class MKLDNNBatchNormHelper implements BatchNormalizationHelper {
}
@Override
public Pair<Gradient, INDArray> backpropGradient(INDArray input, INDArray epsilon, int[] shape, INDArray gamma,
public Pair<Gradient, INDArray> backpropGradient(INDArray input, INDArray epsilon, long[] shape, INDArray gamma,
INDArray dGammaView, INDArray dBetaView, double eps, LayerWorkspaceMgr workspaceMgr) {
//2019-02-14: Backprop disabled pending fixes. https://github.com/deeplearning4j/deeplearning4j/issues/7166
//Also no MKL-DNN implemented for backprop anyway
@ -82,7 +82,7 @@ public class MKLDNNBatchNormHelper implements BatchNormalizationHelper {
}
@Override
public INDArray preOutput(INDArray x, boolean training, int[] shape, INDArray gamma, INDArray beta, INDArray mean, INDArray var,
public INDArray preOutput(INDArray x, boolean training, long[] shape, INDArray gamma, INDArray beta, INDArray mean, INDArray var,
double decay, double eps, LayerWorkspaceMgr workspaceMgr) {
if(x.dataType() != DataType.FLOAT)
return null; //MKL-DNN only supports float

6
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/normalization/BatchNormalization.java
View File

@ -152,10 +152,9 @@ public class BatchNormalization extends BaseLayer<org.deeplearning4j.nn.conf.lay
eps = epsilon;
}
// FIXME: int cast
Pair<Gradient,INDArray> ret = null;
try {
ret = helper.backpropGradient(in, eps, ArrayUtil.toInts(shape), gamma, dGammaView, dBetaView,
ret = helper.backpropGradient(in, eps, shape, gamma, dGammaView, dBetaView,
layerConf.getEps(), workspaceMgr);
} catch (ND4JOpProfilerException e){
throw e; //NaN panic etc for debugging
@ -438,7 +437,6 @@ public class BatchNormalization extends BaseLayer<org.deeplearning4j.nn.conf.lay
//Note that cudnn does not support dense (2d) batch norm case as of v7.1
double decay = layerConf.getDecay();
// FIXME: int cast
INDArray ret = null;
try {
if(globalVarView == null){
@ -448,7 +446,7 @@ public class BatchNormalization extends BaseLayer<org.deeplearning4j.nn.conf.lay
globalVarView.muli(globalVarView);
}
ret = helper.preOutput(in, training == TrainingMode.TRAIN, ArrayUtil.toInts(shape), gamma, beta, globalMeanView,
ret = helper.preOutput(in, training == TrainingMode.TRAIN, shape, gamma, beta, globalMeanView,
globalVarView, decay, layerConf.getEps(), workspaceMgr);
} catch (ND4JOpProfilerException e){
throw e; //NaN panic etc for debugging

4
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/normalization/BatchNormalizationHelper.java
View File

@ -31,10 +31,10 @@ import org.deeplearning4j.nn.workspace.LayerWorkspaceMgr;
public interface BatchNormalizationHelper extends LayerHelper {
boolean checkSupported(double eps, boolean fixedGammaBeta);
Pair<Gradient, INDArray> backpropGradient(INDArray input, INDArray epsilon, int[] shape, INDArray gamma,
Pair<Gradient, INDArray> backpropGradient(INDArray input, INDArray epsilon, long[] shape, INDArray gamma,
INDArray dGammaView, INDArray dBetaView, double eps, LayerWorkspaceMgr workspaceMgr);
INDArray preOutput(INDArray x, boolean training, int[] shape, INDArray gamma, INDArray beta, INDArray mean,
INDArray preOutput(INDArray x, boolean training, long[] shape, INDArray gamma, INDArray beta, INDArray mean,
INDArray var, double decay, double eps, LayerWorkspaceMgr workspaceMgr);
INDArray getMeanCache(DataType dataType);

27
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/objdetect/Yolo2OutputLayer.java
View File

@ -114,10 +114,9 @@ public class Yolo2OutputLayer extends AbstractLayer<org.deeplearning4j.nn.conf.l
double lambdaCoord = layerConf().getLambdaCoord();
double lambdaNoObj = layerConf().getLambdaNoObj();
// FIXME: int cast
int mb = (int) input.size(0);
int h = (int) input.size(2);
int w = (int) input.size(3);
long mb = input.size(0);
long h = input.size(2);
long w = input.size(3);
int b = (int) layerConf().getBoundingBoxes().size(0);
int c = (int) labels.size(1)-4;
@ -243,12 +242,12 @@ public class Yolo2OutputLayer extends AbstractLayer<org.deeplearning4j.nn.conf.l
//Class prediction loss
INDArray classPredictionsPreSoftmax2d = inputClassesPreSoftmax.permute(0,1,3,4,2) //[minibatch, b, c, h, w] To [mb, b, h, w, c]
.dup('c').reshape('c', new int[]{mb*b*h*w, c});
.dup('c').reshape('c', new long[]{mb*b*h*w, c});
INDArray classLabelsBroadcast = Nd4j.createUninitialized(input.dataType(), new long[]{mb, b, c, h, w}, 'c');
for(int i=0; i<b; i++ ){
classLabelsBroadcast.get(all(), point(i), all(), all(), all()).assign(classLabels); //[mb, c, h, w] to [mb, b, c, h, w]
}
INDArray classLabels2d = classLabelsBroadcast.permute(0,1,3,4,2).dup('c').reshape('c', new int[]{mb*b*h*w, c});
INDArray classLabels2d = classLabelsBroadcast.permute(0,1,3,4,2).dup('c').reshape('c', new long[]{mb*b*h*w, c});
//Calculate the loss:
ILossFunction lossConfidence = new LossL2();
@ -297,7 +296,7 @@ public class Yolo2OutputLayer extends AbstractLayer<org.deeplearning4j.nn.conf.l
// ----- Gradient Calculation (specifically: return dL/dIn -----
INDArray epsOut = workspaceMgr.createUninitialized(ArrayType.ACTIVATION_GRAD, input.dataType(), input.shape(), 'c');
INDArray epsOut5 = Shape.newShapeNoCopy(epsOut, new int[]{mb, b, 5+c, h, w}, false);
INDArray epsOut5 = Shape.newShapeNoCopy(epsOut, new long[]{mb, b, 5+c, h, w}, false);
INDArray epsClassPredictions = epsOut5.get(all(), all(), interval(5, 5+c), all(), all()); //Shape: [mb, b, 5+c, h, w]
INDArray epsXY = epsOut5.get(all(), all(), interval(0,2), all(), all());
INDArray epsWH = epsOut5.get(all(), all(), interval(2,4), all(), all());
@ -426,16 +425,16 @@ public class Yolo2OutputLayer extends AbstractLayer<org.deeplearning4j.nn.conf.l
* @return IOU and gradients
*/
private static IOURet calculateIOULabelPredicted(INDArray labelTL, INDArray labelBR, INDArray predictedWH, INDArray predictedXYinGridBox, INDArray objectPresentMask, INDArray objectPresentMaskBool){
// FIXME: int cast
int mb = (int) labelTL.size(0);
int h = (int) labelTL.size(2);
int w = (int) labelTL.size(3);
int b = (int) predictedWH.size(1);
long mb = labelTL.size(0);
long h = labelTL.size(2);
long w = labelTL.size(3);
long b = predictedWH.size(1);
INDArray labelWH = labelBR.sub(labelTL); //4d [mb, 2, H, W], label W/H in terms of number of grid boxes
int gridH = (int) labelTL.size(2);
int gridW = (int) labelTL.size(3);
long gridH = labelTL.size(2);
long gridW = labelTL.size(3);
//Add grid positions to the predicted XY values (to get predicted XY in terms of grid cell units in image,
// from (0 to 1 in grid cell) format)
INDArray linspaceX = Nd4j.linspace(0, gridW-1, gridW, predictedWH.dataType());

24
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/objdetect/YoloUtils.java
View File

@ -45,12 +45,11 @@ public class YoloUtils {
}
public static INDArray activate(@NonNull INDArray boundingBoxPriors, @NonNull INDArray input, LayerWorkspaceMgr layerWorkspaceMgr){
// FIXME: int cast
int mb = (int) input.size(0);
int h = (int) input.size(2);
int w = (int) input.size(3);
int b = (int) boundingBoxPriors.size(0);
int c = (int) (input.size(1)/b)-5; //input.size(1) == b * (5 + C) -> C = (input.size(1)/b) - 5
long mb = input.size(0);
long h = input.size(2);
long w = input.size(3);
long b = boundingBoxPriors.size(0);
long c = input.size(1)/b-5; //input.size(1) == b * (5 + C) -> C = (input.size(1)/b) - 5
INDArray output = layerWorkspaceMgr.create(ArrayType.ACTIVATIONS, input.dataType(), input.shape(), 'c');
INDArray output5 = output.reshape('c', mb, b, 5+c, h, w);
@ -77,7 +76,7 @@ public class YoloUtils {
//TODO OPTIMIZE?
INDArray inputClassesPreSoftmax = input5.get(all(), all(), interval(5, 5+c), all(), all()); //Shape: [minibatch, C, H, W]
INDArray classPredictionsPreSoftmax2d = inputClassesPreSoftmax.permute(0,1,3,4,2) //[minibatch, b, c, h, w] To [mb, b, h, w, c]
.dup('c').reshape('c', new int[]{mb*b*h*w, c});
.dup('c').reshape('c', new long[]{mb*b*h*w, c});
Transforms.softmax(classPredictionsPreSoftmax2d, false);
INDArray postSoftmax5d = classPredictionsPreSoftmax2d.reshape('c', mb, b, h, w, c ).permute(0, 1, 4, 2, 3);
@ -173,13 +172,12 @@ public class YoloUtils {
throw new IllegalStateException("Invalid confidence threshold: must be in range [0,1]. Got: " + confThreshold);
}
// FIXME: int cast
//Activations format: [mb, 5b+c, h, w]
int mb = (int) networkOutput.size(0);
int h = (int) networkOutput.size(2);
int w = (int) networkOutput.size(3);
int b = (int) boundingBoxPriors.size(0);
int c = (int) (networkOutput.size(1)/b)-5; //input.size(1) == b * (5 + C) -> C = (input.size(1)/b) - 5
long mb = networkOutput.size(0);
long h = networkOutput.size(2);
long w = networkOutput.size(3);
long b = boundingBoxPriors.size(0);
long c = (networkOutput.size(1)/b)-5; //input.size(1) == b * (5 + C) -> C = (input.size(1)/b) - 5
//Reshape from [minibatch, B*(5+C), H, W] to [minibatch, B, 5+C, H, W] to [minibatch, B, 5, H, W]
INDArray output5 = networkOutput.dup('c').reshape(mb, b, 5+c, h, w);

10
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/recurrent/LSTMHelpers.java
View File

@ -40,6 +40,7 @@ import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.api.ops.impl.transforms.pairwise.arithmetic.MulOp;
import org.nd4j.linalg.api.ops.impl.transforms.same.TimesOneMinus;
import org.nd4j.linalg.api.shape.Shape;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.exception.ND4JOpProfilerException;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.indexing.NDArrayIndex;
@ -113,7 +114,9 @@ public class LSTMHelpers {
input = input.castTo(inputWeights.dataType()); //No-op if already correct dtype
// FIXME
if ((!is2dInput && (input.size(2) > Integer.MAX_VALUE)) ||
recurrentWeights.size(0) > Integer.MAX_VALUE || input.size(0) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
int timeSeriesLength = (int) (is2dInput ? 1 : input.size(2));
int hiddenLayerSize = (int) recurrentWeights.size(0);
int miniBatchSize = (int) input.size(0);
@ -550,7 +553,8 @@ public class LSTMHelpers {
for (long iTimeIndex = timeSeriesLength - 1; iTimeIndex >= endIdx; iTimeIndex--) {
try(MemoryWorkspace ws = workspaceMgr.notifyScopeEntered(ArrayType.RNN_BP_LOOP_WORKING_MEM)) {
// FIXME: int cast
if (iTimeIndex > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
int time = (int) iTimeIndex;
int inext = 1;
@ -574,8 +578,6 @@ public class LSTMHelpers {
(iTimeIndex == 0 ? fwdPass.prevAct : fwdPass.fwdPassOutputAsArrays[(int) (time - inext)]);
INDArray currMemCellState = fwdPass.memCellState[(int) time];
// FIXME: int cast
//LSTM unit output errors (dL/d(a_out)); not to be confused with \delta=dL/d(z_out)
INDArray epsilonSlice = (is2dInput ? epsilon : epsilon.tensorAlongDimension((int) time, 1, 0)); //(w^{L+1}*(delta^{(L+1)t})^T)^T or equiv.

7
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/recurrent/RnnLossLayer.java
View File

@ -89,8 +89,7 @@ public class RnnLossLayer extends BaseLayer<org.deeplearning4j.nn.conf.layers.Rn
ILossFunction lossFunction = layerConf().getLossFn();
INDArray delta2d = lossFunction.computeGradient(labels2d, input2d.dup(input2d.ordering()), layerConf().getActivationFn(), maskReshaped);
// FIXME: int cast
INDArray delta3d = TimeSeriesUtils.reshape2dTo3d(delta2d, (int) input.size(0), workspaceMgr, ArrayType.ACTIVATION_GRAD);
INDArray delta3d = TimeSeriesUtils.reshape2dTo3d(delta2d, input.size(0), workspaceMgr, ArrayType.ACTIVATION_GRAD);
// grab the empty gradient
Gradient gradient = new DefaultGradient();
@ -119,7 +118,6 @@ public class RnnLossLayer extends BaseLayer<org.deeplearning4j.nn.conf.layers.Rn
@Override
public int numLabels() {
// FIXME: int cast
return (int) labels.size(1);
}
@ -167,7 +165,7 @@ public class RnnLossLayer extends BaseLayer<org.deeplearning4j.nn.conf.layers.Rn
INDArray as2d = TimeSeriesUtils.reshape3dTo2d(input);
INDArray out2d = layerConf().getActivationFn().getActivation(workspaceMgr.dup(ArrayType.ACTIVATIONS, as2d, as2d.ordering()), training);
return workspaceMgr.leverageTo(ArrayType.ACTIVATIONS, TimeSeriesUtils.reshape2dTo3d(out2d, (int)input.size(0), workspaceMgr, ArrayType.ACTIVATIONS));
return workspaceMgr.leverageTo(ArrayType.ACTIVATIONS, TimeSeriesUtils.reshape2dTo3d(out2d, input.size(0), workspaceMgr, ArrayType.ACTIVATIONS));
}
@Override
@ -254,7 +252,6 @@ public class RnnLossLayer extends BaseLayer<org.deeplearning4j.nn.conf.layers.Rn
//scoreArray: shape [minibatch*timeSeriesLength, 1]
//Reshape it to [minibatch, timeSeriesLength] then sum over time step
// FIXME: int cast
INDArray scoreArrayTs = TimeSeriesUtils.reshapeVectorToTimeSeriesMask(scoreArray, (int)input.size(0));
INDArray summedScores = scoreArrayTs.sum(1);

7
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/recurrent/RnnOutputLayer.java
View File

@ -70,8 +70,7 @@ public class RnnOutputLayer extends BaseOutputLayer<org.deeplearning4j.nn.conf.l
this.input = inputTemp;
INDArray epsilon2d = gradAndEpsilonNext.getSecond();
// FIXME: int cast
INDArray epsilon3d = TimeSeriesUtils.reshape2dTo3d(epsilon2d, (int) input.size(0), workspaceMgr, ArrayType.ACTIVATION_GRAD);
INDArray epsilon3d = TimeSeriesUtils.reshape2dTo3d(epsilon2d, input.size(0), workspaceMgr, ArrayType.ACTIVATION_GRAD);
weightNoiseParams.clear();
@ -145,8 +144,7 @@ public class RnnOutputLayer extends BaseOutputLayer<org.deeplearning4j.nn.conf.l
}
}
// FIXME: int cast
return TimeSeriesUtils.reshape2dTo3d(act2d, (int) input.size(0), workspaceMgr, ArrayType.ACTIVATIONS);
return TimeSeriesUtils.reshape2dTo3d(act2d, input.size(0), workspaceMgr, ArrayType.ACTIVATIONS);
}
@Override
@ -205,7 +203,6 @@ public class RnnOutputLayer extends BaseOutputLayer<org.deeplearning4j.nn.conf.l
//scoreArray: shape [minibatch*timeSeriesLength, 1]
//Reshape it to [minibatch, timeSeriesLength] then sum over time step
// FIXME: int cast
INDArray scoreArrayTs = TimeSeriesUtils.reshapeVectorToTimeSeriesMask(scoreArray, (int)input.size(0));
INDArray summedScores = scoreArrayTs.sum(true, 1);

7
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/layers/variational/VariationalAutoencoder.java
View File

@ -41,6 +41,7 @@ import org.nd4j.linalg.api.blas.Level1;
import org.nd4j.linalg.api.buffer.DataType;
import org.nd4j.linalg.api.memory.MemoryWorkspace;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.learning.regularization.Regularization;
import org.nd4j.linalg.lossfunctions.ILossFunction;
@ -552,7 +553,8 @@ public class VariationalAutoencoder implements Layer {
@Override
public int batchSize() {
// FIXME: int cast
if (input.size(0) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
return (int) input.size(0);
}
@ -862,7 +864,8 @@ public class VariationalAutoencoder implements Layer {
@Override
public int getInputMiniBatchSize() {
// FIXME: int cast
if (input.size(0) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
return (int) input.size(0);
}

41
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/multilayer/MultiLayerNetwork.java
View File

@ -75,6 +75,7 @@ import org.nd4j.linalg.dataset.DataSet;
import org.nd4j.linalg.dataset.api.MultiDataSet;
import org.nd4j.linalg.dataset.api.iterator.DataSetIterator;
import org.nd4j.linalg.dataset.api.iterator.MultiDataSetIterator;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.heartbeat.Heartbeat;
import org.nd4j.linalg.heartbeat.reports.Environment;
@ -425,7 +426,8 @@ public class MultiLayerNetwork implements Serializable, Classifier, Layer, Neura
try(MemoryWorkspace ws = workspaceMgr.notifyScopeEntered(ArrayType.FF_WORKING_MEM)) {
if (layerWiseConfigurations.getInputPreProcess(layerIdx) != null) {
// FIXME: int cast
if (input.size(0) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
outputOfPrevLayer = layerWiseConfigurations.getInputPreProcess(layerIdx).preProcess(outputOfPrevLayer, (int) input.size(0),
LayerWorkspaceMgr.noWorkspaces(helperWorkspaces));
}
@ -439,7 +441,8 @@ public class MultiLayerNetwork implements Serializable, Classifier, Layer, Neura
//In 99+% of cases, the input and labels dimension 0 size should be identical
//The only real exceptions: space to batch, and batch to space layers
//In those cases, we should base it on the labels size, as this impacts gradient calculation
// FIXME: int cast
if (input.size(0) > Integer.MAX_VALUE || labels.size(0) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
return labels == null ? (int) input.size(0) : (int)labels.size(0);
}
@ -2074,7 +2077,8 @@ public class MultiLayerNetwork implements Serializable, Classifier, Layer, Neura
if (endTimeIdx > timeSeriesLength)
endTimeIdx = timeSeriesLength;
// FIXME: int cast
if (startTimeIdx > Integer.MAX_VALUE || endTimeIdx > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
INDArray[] subsets = getSubsetsForTbptt((int) startTimeIdx, (int) endTimeIdx, input, labels,
featuresMaskArray, labelsMaskArray);
@ -2211,7 +2215,9 @@ public class MultiLayerNetwork implements Serializable, Classifier, Layer, Neura
public int[] predict(INDArray d) {
INDArray output = output(d, Layer.TrainingMode.TEST);
// FIXME: int cast
if (d.size(0) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
int[] ret = new int[(int) d.size(0)];
if (d.isRowVectorOrScalar())
ret[0] = Nd4j.getBlasWrapper().iamax(output);
@ -2335,7 +2341,8 @@ public class MultiLayerNetwork implements Serializable, Classifier, Layer, Neura
org.deeplearning4j.nn.conf.layers.OutputLayer layerConf =
(org.deeplearning4j.nn.conf.layers.OutputLayer) getOutputLayer().conf().getLayer();
// FIXME: int cast
if (layerConf.getNOut() > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
fit(examples, FeatureUtil.toOutcomeMatrix(labels, (int) layerConf.getNOut()));
}
@ -2584,7 +2591,8 @@ public class MultiLayerNetwork implements Serializable, Classifier, Layer, Neura
INDArray inputToOutputLayer = outputOfLayerDetached(training, FwdPassType.STANDARD,layers.length-2, data.getFeatures(),
data.getFeaturesMaskArray(), data.getLabelsMaskArray(), null);
// FIXME: int cast
if (data.getFeatures().size(0) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
IOutputLayer ol = (IOutputLayer) getOutputLayer();
if (getLayerWiseConfigurations().getInputPreProcess(layers.length - 1) != null) {
inputToOutputLayer = getLayerWiseConfigurations().getInputPreProcess(layers.length - 1)
@ -2647,7 +2655,8 @@ public class MultiLayerNetwork implements Serializable, Classifier, Layer, Neura
IOutputLayer ol = (IOutputLayer) getOutputLayer();
if(layerWiseConfigurations.getInputPreProcess(layers.length-1) != null){
// FIXME: int cast
if (data.getFeatures().size(0) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
inputLast = layerWiseConfigurations.getInputPreProcess(layers.length-1).preProcess(inputLast,
(int) data.getFeatures().size(0), mgr);
}
@ -2811,7 +2820,8 @@ public class MultiLayerNetwork implements Serializable, Classifier, Layer, Neura
throw new IllegalArgumentException(
"Invalid input: length 0 (shape: " + Arrays.toString(input.shape()) + ")");
// FIXME: int cast
if (input.size(0) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
setInputMiniBatchSize((int) input.size(0));
}
}
@ -3086,7 +3096,8 @@ public class MultiLayerNetwork implements Serializable, Classifier, Layer, Neura
if(!conf().isMiniBatch())
return 1;
// FIXME: int cast
if (input.size(0) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
return (int) input.size(0);
}
@ -3256,7 +3267,8 @@ public class MultiLayerNetwork implements Serializable, Classifier, Layer, Neura
public void setLayerMaskArrays(INDArray featuresMaskArray, INDArray labelsMaskArray) {
if (featuresMaskArray != null) {
// FIXME: int cast
if (featuresMaskArray.size(0) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
//New approach: use feedForwardMaskArray method
feedForwardMaskArray(featuresMaskArray, MaskState.Active, (int) featuresMaskArray.size(0));
@ -3438,7 +3450,8 @@ public class MultiLayerNetwork implements Serializable, Classifier, Layer, Neura
val startTimeIdx = i * fwdLen;
val endTimeIdx = Math.min(startTimeIdx + fwdLen, tsLength);
// FIXME: int cast
if (endTimeIdx > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
INDArray[] subsets = getSubsetsForTbptt(startTimeIdx, (int) endTimeIdx, features, labels, fMask, lMask);
setLayerMaskArrays(subsets[2], subsets[3]);
@ -3943,7 +3956,8 @@ public class MultiLayerNetwork implements Serializable, Classifier, Layer, Neura
}
FeedForwardLayer ffl = (FeedForwardLayer) conf;
// FIXME: int cast
if (ffl.getNOut() > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
return (int) ffl.getNOut();
}
@ -3969,7 +3983,8 @@ public class MultiLayerNetwork implements Serializable, Classifier, Layer, Neura
}
FeedForwardLayer ffl = (FeedForwardLayer) conf;
// FIXME: int cast
if (ffl.getNIn() > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
return (int) ffl.getNIn();
}

10
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/params/VariationalAutoencoderParamInitializer.java
View File

@ -22,6 +22,7 @@ import org.deeplearning4j.nn.conf.layers.Layer;
import org.deeplearning4j.nn.conf.layers.variational.VariationalAutoencoder;
import org.deeplearning4j.nn.weights.IWeightInit;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.indexing.NDArrayIndex;
import java.util.ArrayList;
@ -108,7 +109,8 @@ public class VariationalAutoencoderParamInitializer extends DefaultParamInitiali
}
//Between last decoder layer and parameters for p(x|z):
// FIXME: int cast
if (nIn > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
val nDistributionParams = layer.getOutputDistribution().distributionInputSize((int) nIn);
val lastDecLayerSize = decoderLayerSizes[decoderLayerSizes.length - 1];
paramCount += (lastDecLayerSize + 1) * nDistributionParams;
@ -294,7 +296,8 @@ public class VariationalAutoencoderParamInitializer extends DefaultParamInitiali
}
//Finally, p(x|z):
// FIXME: int cast
if (nIn > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
int nDistributionParams = layer.getOutputDistribution().distributionInputSize((int) nIn);
int pxzWeightCount = decoderLayerSizes[decoderLayerSizes.length - 1] * nDistributionParams;
INDArray pxzWeightView =
@ -402,7 +405,8 @@ public class VariationalAutoencoderParamInitializer extends DefaultParamInitiali
}
//Finally, p(x|z):
// FIXME: int cast
if (nIn > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
int nDistributionParams = layer.getOutputDistribution().distributionInputSize((int) nIn);
int pxzWeightCount = decoderLayerSizes[decoderLayerSizes.length - 1] * nDistributionParams;
INDArray pxzWeightView =

10
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/nn/updater/BaseMultiLayerUpdater.java
View File

@ -30,6 +30,7 @@ import org.nd4j.linalg.api.ops.CustomOp;
import org.nd4j.linalg.api.ops.DynamicCustomOp;
import org.nd4j.linalg.api.ops.impl.reduce.floating.Norm2;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.indexing.NDArrayIndex;
import org.deeplearning4j.nn.workspace.ArrayType;
@ -111,7 +112,8 @@ public abstract class BaseMultiLayerUpdater<T extends Model> implements Updater
if (currentBlock == null || !UpdaterUtils.updaterConfigurationsEquals(lastLayer, lastVariable,
layers[i], var)) {
// FIXME: int cast
if (paramsViewSoFar + paramSizeThisVariable > Integer.MAX_VALUE || paramsViewSoFar + paramSizeThisVariable > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
//Create a new block
List<UpdaterBlock.ParamState> list = new ArrayList<>();
list.add(new UpdaterBlock.ParamState(layers[i], var, paramsViewSoFar,
@ -122,9 +124,11 @@ public abstract class BaseMultiLayerUpdater<T extends Model> implements Updater
updaterBlocks.add(currentBlock);
} else {
// FIXME: int cast
long newOffset = currentBlock.getParamOffsetEnd() + paramSizeThisVariable;
if (newOffset > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
//Add to existing updater block
currentBlock.setParamOffsetEnd((int) (currentBlock.getParamOffsetEnd() + paramSizeThisVariable));
currentBlock.setParamOffsetEnd((int) newOffset);
currentBlock.setUpdaterViewOffsetEnd(
currentBlock.getUpdaterViewOffsetEnd() + updaterStateSizeThisVariable);
currentBlock.getLayersAndVariablesInBlock()

9
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/util/Convolution1DUtils.java
View File

@ -29,6 +29,7 @@ import org.nd4j.base.Preconditions;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.api.ops.impl.broadcast.BroadcastCopyOp;
import org.nd4j.linalg.api.shape.Shape;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.factory.Nd4j;
import java.util.Arrays;
@ -62,10 +63,9 @@ public class Convolution1DUtils {
* @param dilation Kernel dilation
* @return Output size (width)
*/
public static int getOutputSize(int inH, int kernel, int strides, int padding,
public static long getOutputSize(long inH, int kernel, int strides, int padding,
ConvolutionMode convolutionMode, int dilation) {
// FIXME: int cast
int eKernel = effectiveKernelSize(kernel, dilation);
long eKernel = effectiveKernelSize(kernel, dilation);
if (convolutionMode == ConvolutionMode.Same) {
return (int) Math.ceil(inH / ((double) strides));
}
@ -85,7 +85,8 @@ public class Convolution1DUtils {
*/
public static int getOutputSize(INDArray inputData, int kernel, int strides, int padding,
ConvolutionMode convolutionMode, int dilation) {
// FIXME: int cast
if (inputData.size(2) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
int inH = (int) inputData.size(2);
int eKernel = effectiveKernelSize(kernel, dilation);
boolean atrous = (eKernel == kernel);

17
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/util/Convolution3DUtils.java
View File

@ -61,15 +61,14 @@ public class Convolution3DUtils {
ConvolutionMode convolutionMode, int[] dilation, boolean isNCDHW) {
// NCDHW vs. NDHWC
int inD = (int) (isNCDHW ? inputData.size(2) : inputData.size(1));
int inH = (int) (isNCDHW ? inputData.size(3) : inputData.size(2));
int inW = (int) (isNCDHW ? inputData.size(4) : inputData.size(3));
long inD = (isNCDHW ? inputData.size(2) : inputData.size(1));
long inH = (isNCDHW ? inputData.size(3) : inputData.size(2));
long inW = (isNCDHW ? inputData.size(4) : inputData.size(3));
int[] eKernel = effectiveKernelSize(kernel, dilation);
boolean atrous = (eKernel == kernel);
// FIXME: int cast
val inShape = new int[]{inD, inH, inW};
val inShape = new long[]{inD, inH, inW};
validateShapes(ArrayUtil.toInts(inputData.shape()), eKernel, strides, padding, convolutionMode, dilation, inShape, atrous);
if (convolutionMode == ConvolutionMode.Same) {
@ -80,16 +79,16 @@ public class Convolution3DUtils {
return new int[]{outD, outH, outW};
}
int outD = (inD - eKernel[0] + 2 * padding[0]) / strides[0] + 1;
int outH = (inH - eKernel[1] + 2 * padding[1]) / strides[1] + 1;
int outW = (inW - eKernel[2] + 2 * padding[2]) / strides[2] + 1;
int outD = ((int)inD - eKernel[0] + 2 * padding[0]) / strides[0] + 1;
int outH = ((int)inH - eKernel[1] + 2 * padding[1]) / strides[1] + 1;
int outW = ((int)inW - eKernel[2] + 2 * padding[2]) / strides[2] + 1;
return new int[]{outD, outH, outW};
}
private static void validateShapes(int[] inputDataShape, int[] eKernel, int[] strides, int[] padding,
ConvolutionMode convolutionMode, int[] dilation, int[] inShape,
ConvolutionMode convolutionMode, int[] dilation, long[] inShape,
boolean atrous) {
String[] dims = new String[]{"depth", "height", "width"};

21
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/util/ConvolutionUtils.java
View File

@ -36,6 +36,8 @@ import org.nd4j.linalg.api.ops.impl.broadcast.BroadcastCopyOp;
import org.nd4j.linalg.api.ops.impl.layers.convolution.MaxPooling2D;
import org.nd4j.linalg.api.ops.impl.layers.convolution.config.Pooling2DConfig;
import org.nd4j.linalg.api.shape.Shape;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.factory.NDArrayFactory;
import org.nd4j.linalg.factory.Nd4j;
import java.util.Arrays;
@ -73,7 +75,8 @@ public class ConvolutionUtils {
public static int[] getDeconvolutionOutputSize(INDArray inputData, int[] kernel, int[] strides, int[] padding,
ConvolutionMode convolutionMode, int[] dilation) {
// FIXME: int cast
if (inputData.size(2) > Integer.MAX_VALUE || inputData.size(3) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
int hIn = (int) inputData.size(2);
int wIn = (int) inputData.size(3);
int[] eKernel = effectiveKernelSize(kernel, dilation);
@ -104,7 +107,8 @@ public class ConvolutionUtils {
*/
public static int[] getOutputSize(INDArray inputData, int[] kernel, int[] strides, int[] padding,
ConvolutionMode convolutionMode, int[] dilation) {
// FIXME: int cast
if (inputData.size(2) > Integer.MAX_VALUE || inputData.size(3) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
int inH = (int) inputData.size(2);
int inW = (int) inputData.size(3);
@ -499,7 +503,7 @@ public class ConvolutionUtils {
}
}
public static INDArray reshape2dTo4d(INDArray in2d, int[] toShape, LayerWorkspaceMgr workspaceMgr, ArrayType type){
public static INDArray reshape2dTo4d(INDArray in2d, long[] toShape, LayerWorkspaceMgr workspaceMgr, ArrayType type){
if(in2d.rank() != 2)
throw new IllegalArgumentException("Invalid input: expect NDArray with rank 2");
if (toShape.length != 4)
@ -513,7 +517,7 @@ public class ConvolutionUtils {
return workspaceMgr.leverageTo(type, out.permute(0, 3, 1, 2));
}
public static INDArray reshape2dTo5d(Convolution3D.DataFormat format, INDArray in2d, int n, int d, int h, int w, int ch, LayerWorkspaceMgr workspaceMgr, ArrayType type){
public static INDArray reshape2dTo5d(Convolution3D.DataFormat format, INDArray in2d, long n, long d, long h, long w, long ch, LayerWorkspaceMgr workspaceMgr, ArrayType type){
if(in2d.rank() != 2)
throw new IllegalArgumentException("Invalid input: expect NDArray with rank 2");
@ -580,14 +584,21 @@ public class ConvolutionUtils {
int inW;
int inDepth;
// FIXME: int cast
if (inputType instanceof InputType.InputTypeConvolutional) {
InputType.InputTypeConvolutional conv = (InputType.InputTypeConvolutional) inputType;
if (conv.getHeight() > Integer.MAX_VALUE || conv.getWidth() > Integer.MAX_VALUE ||
conv.getChannels() > Integer.MAX_VALUE){
throw new ND4JArraySizeException();
}
inH = (int) conv.getHeight();
inW = (int) conv.getWidth();
inDepth = (int) conv.getChannels();
} else if (inputType instanceof InputType.InputTypeConvolutionalFlat) {
InputType.InputTypeConvolutionalFlat conv = (InputType.InputTypeConvolutionalFlat) inputType;
if (conv.getHeight() > Integer.MAX_VALUE || conv.getWidth() > Integer.MAX_VALUE ||
conv.getDepth() > Integer.MAX_VALUE) {
throw new ND4JArraySizeException();
}
inH = (int) conv.getHeight();
inW = (int) conv.getWidth();
inDepth = (int) conv.getDepth();

19
deeplearning4j/deeplearning4j-nn/src/main/java/org/deeplearning4j/util/TimeSeriesUtils.java
View File

@ -20,6 +20,7 @@ import lombok.val;
import org.nd4j.base.Preconditions;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.api.shape.Shape;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.indexing.BooleanIndexing;
import org.nd4j.linalg.indexing.INDArrayIndex;
@ -193,7 +194,7 @@ public class TimeSeriesUtils {
}
public static INDArray reshape2dTo3d(INDArray in, int miniBatchSize, LayerWorkspaceMgr workspaceMgr, ArrayType arrayType) {
public static INDArray reshape2dTo3d(INDArray in, long miniBatchSize, LayerWorkspaceMgr workspaceMgr, ArrayType arrayType) {
if (in.rank() != 2)
throw new IllegalArgumentException("Invalid input: expect NDArray with rank 2");
//Based on: RnnToFeedForwardPreProcessor
@ -220,7 +221,6 @@ public class TimeSeriesUtils {
in = in.dup('f');
}
// FIXME: int cast
int[] idxs = new int[(int) in.size(2)];
int j=0;
for( int i=idxs.length-1; i>=0; i--){
@ -248,7 +248,8 @@ public class TimeSeriesUtils {
in = workspaceMgr.dup(arrayType, in, 'f');
}
// FIXME: int cast
if (in.size(2) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
int[] idxs = new int[(int) in.size(2)];
int j=0;
for( int i=idxs.length-1; i>=0; i--){
@ -291,7 +292,8 @@ public class TimeSeriesUtils {
+ " with shape " + Arrays.toString(mask.shape()));
}
// FIXME: int cast
if (mask.size(1) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
int[] idxs = new int[(int) mask.size(1)];
int j=0;
for( int i=idxs.length-1; i>=0; i--){
@ -319,7 +321,8 @@ public class TimeSeriesUtils {
+ " with shape " + Arrays.toString(mask.shape()));
}
// FIXME: int cast
if (mask.size(1) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
int[] idxs = new int[(int) mask.size(1)];
int j=0;
for( int i=idxs.length-1; i>=0; i--){
@ -358,9 +361,8 @@ public class TimeSeriesUtils {
INDArray out;
if (mask == null) {
// FIXME: int cast
//No mask array -> extract same (last) column for all
int lastTS = (int) pullFrom.size(2) - 1;
long lastTS = pullFrom.size(2) - 1;
out = pullFrom.get(NDArrayIndex.all(), NDArrayIndex.all(), NDArrayIndex.point(lastTS));
fwdPassTimeSteps = null; //Null -> last time step for all examples
} else {
@ -396,9 +398,8 @@ public class TimeSeriesUtils {
INDArray out;
if (mask == null) {
// FIXME: int cast
//No mask array -> extract same (last) column for all
int lastTS = (int) pullFrom.size(2) - 1;
long lastTS = pullFrom.size(2) - 1;
out = pullFrom.get(NDArrayIndex.all(), NDArrayIndex.all(), NDArrayIndex.point(lastTS));
fwdPassTimeSteps = null; //Null -> last time step for all examples
} else {

1
deeplearning4j/deeplearning4j-nn/src/test/java/org/deeplearning4j/TestUtils.java
View File

@ -116,7 +116,6 @@ public class TestUtils {
public static INDArray randomOneHot(long examples, long nOut, Random rng){
INDArray arr = Nd4j.create(examples, nOut);
for( int i=0; i<examples; i++ ){
// FIXME: int cast
arr.putScalar(i, rng.nextInt((int) nOut), 1.0);
}
return arr;

1
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark-nlp/src/main/java/org/deeplearning4j/spark/models/embeddings/word2vec/FirstIterationFunction.java
View File

@ -214,7 +214,6 @@ public class FirstIterationFunction implements
else {
nextRandom.set(Math.abs(nextRandom.get() * 25214903917L + 11));
// FIXME: int cast
int idx = Math.abs((int) (nextRandom.get() >> 16) % (int) negativeHolder.getTable().length());
target = negativeHolder.getTable().getInt(idx);

1
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark-nlp/src/main/java/org/deeplearning4j/spark/models/embeddings/word2vec/SecondIterationFunction.java
View File

@ -222,7 +222,6 @@ public class SecondIterationFunction implements FlatMapFunction<Iterator<Tuple2<
else {
nextRandom.set(Math.abs(nextRandom.get() * 25214903917L + 11));
// FIXME: int cast
int idx = (int) Math.abs((int) (nextRandom.get() >> 16) % negativeHolder.getTable().length());
target = negativeHolder.getTable().getInt(idx);

1
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark-nlp/src/main/java/org/deeplearning4j/spark/models/embeddings/word2vec/SentenceBatch.java
View File

@ -162,7 +162,6 @@ public class SentenceBatch implements Function<Word2VecFuncCall, Word2VecChange>
label = 1;
} else {
nextRandom.set(nextRandom.get() * 25214903917L + 11);
// FIXME: int cast
target = table.getInt((int) (nextRandom.get() >> 16) % (int) table.length());
if (target == 0)
target = (int) nextRandom.get() % (numWords - 1) + 1;

1
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark-nlp/src/main/java/org/deeplearning4j/spark/models/embeddings/word2vec/Word2VecPerformer.java
View File

@ -187,7 +187,6 @@ public class Word2VecPerformer implements VoidFunction<Pair<List<VocabWord>, Ato
} else {
nextRandom.set(nextRandom.get() * 25214903917L + 11);
// FIXME: int cast
target = table.getInt((int) (nextRandom.get() >> 16) % (int) table.length());
if (target == 0)
target = (int) nextRandom.get() % (numWords - 1) + 1;

1
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark-nlp/src/main/java/org/deeplearning4j/spark/models/embeddings/word2vec/Word2VecPerformerVoid.java
View File

@ -337,7 +337,6 @@ public class Word2VecPerformerVoid implements VoidFunction<Pair<List<VocabWord>,
label = 1;
} else {
nextRandom.set(nextRandom.get() * 25214903917L + 11);
// FIXME: int cast
target = table.getInt((int) (nextRandom.get() >> 16) % (int) table.length());
if (target == 0)
target = (int) nextRandom.get() % (numWords - 1) + 1;

4
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark/src/main/java/org/deeplearning4j/spark/api/stats/StatsCalculationHelper.java
View File

@ -39,7 +39,7 @@ public class StatsCalculationHelper {
private long initialModelAfter;
private long lastDataSetBefore;
private long lastProcessBefore;
private int totalExampleCount;
private long totalExampleCount;
private List<EventStats> dataSetGetTimes = new ArrayList<>();
private List<EventStats> processMiniBatchTimes = new ArrayList<>();
@ -65,7 +65,7 @@ public class StatsCalculationHelper {
lastDataSetBefore = timeSource.currentTimeMillis();
}
public void logNextDataSetAfter(int numExamples) {
public void logNextDataSetAfter(long numExamples) {
long now = timeSource.currentTimeMillis();
long duration = now - lastDataSetBefore;
dataSetGetTimes.add(new BaseEventStats(lastDataSetBefore, duration));

3
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark/src/main/java/org/deeplearning4j/spark/api/worker/ExecuteWorkerMultiDataSetFlatMap.java
View File

@ -84,9 +84,8 @@ public class ExecuteWorkerMultiDataSetFlatMap<R extends TrainingResult> implemen
s.logNextDataSetBefore();
MultiDataSet next = batchedIterator.next();
// FIXME: int cast
if (stats)
s.logNextDataSetAfter((int) next.getFeatures(0).size(0));
s.logNextDataSetAfter(next.getFeatures(0).size(0));
if (stats) {
s.logProcessMinibatchBefore();

9
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark/src/main/java/org/deeplearning4j/spark/impl/graph/scoring/GraphFeedForwardWithKeyFunction.java
View File

@ -71,7 +71,7 @@ public class GraphFeedForwardWithKeyFunction<K> implements PairFlatMapFunction<I
List<INDArray[]> featuresList = new ArrayList<>(batchSize);
List<K> keyList = new ArrayList<>(batchSize);
List<Integer> origSizeList = new ArrayList<>();
List<Long> origSizeList = new ArrayList<>();
long[][] firstShapes = null;
boolean sizesDiffer = false;
@ -96,8 +96,7 @@ public class GraphFeedForwardWithKeyFunction<K> implements PairFlatMapFunction<I
featuresList.add(t2._2());
keyList.add(t2._1());
// FIXME: int cast
origSizeList.add((int) t2._2()[0].size(0));
origSizeList.add(t2._2()[0].size(0));
tupleCount++;
}
@ -156,7 +155,7 @@ public class GraphFeedForwardWithKeyFunction<K> implements PairFlatMapFunction<I
examplesInBatch = 0;
for (int i = firstIdx; i < nextIdx; i++) {
int numExamples = origSizeList.get(i);
long numExamples = origSizeList.get(i);
INDArray[] outSubset = new INDArray[out.length];
for (int j = 0; j < out.length; j++) {
outSubset[j] = getSubset(examplesInBatch, examplesInBatch + numExamples, out[j]);
@ -174,7 +173,7 @@ public class GraphFeedForwardWithKeyFunction<K> implements PairFlatMapFunction<I
return output.iterator();
}
private INDArray getSubset(int exampleStart, int exampleEnd, INDArray from) {
private INDArray getSubset(long exampleStart, long exampleEnd, INDArray from) {
switch (from.rank()) {
case 2:
return from.get(NDArrayIndex.interval(exampleStart, exampleEnd), NDArrayIndex.all());

11
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark/src/main/java/org/deeplearning4j/spark/impl/graph/scoring/ScoreFlatMapFunctionCGDataSet.java
View File

@ -36,7 +36,7 @@ import java.util.Iterator;
import java.util.List;
/** Function used to score a DataSet using a ComputationGraph */
public class ScoreFlatMapFunctionCGDataSet implements FlatMapFunction<Iterator<DataSet>, Tuple2<Integer, Double>> {
public class ScoreFlatMapFunctionCGDataSet implements FlatMapFunction<Iterator<DataSet>, Tuple2<Long, Double>> {
private static final Logger log = LoggerFactory.getLogger(ScoreFlatMapFunctionCGDataSet.class);
private String json;
private Broadcast<INDArray> params;
@ -50,9 +50,9 @@ public class ScoreFlatMapFunctionCGDataSet implements FlatMapFunction<Iterator<D
}
@Override
public Iterator<Tuple2<Integer, Double>> call(Iterator<DataSet> dataSetIterator) throws Exception {
public Iterator<Tuple2<Long, Double>> call(Iterator<DataSet> dataSetIterator) throws Exception {
if (!dataSetIterator.hasNext()) {
return Collections.singletonList(new Tuple2<>(0, 0.0)).iterator();
return Collections.singletonList(new Tuple2<>(0L, 0.0)).iterator();
}
DataSetIterator iter = new IteratorDataSetIterator(dataSetIterator, minibatchSize); //Does batching where appropriate
@ -65,13 +65,12 @@ public class ScoreFlatMapFunctionCGDataSet implements FlatMapFunction<Iterator<D
"Network did not have same number of parameters as the broadcast set parameters");
network.setParams(val);
List<Tuple2<Integer, Double>> out = new ArrayList<>();
List<Tuple2<Long, Double>> out = new ArrayList<>();
while (iter.hasNext()) {
DataSet ds = iter.next();
double score = network.score(ds, false);
// FIXME: int cast
int numExamples = (int) ds.getFeatures().size(0);
long numExamples = ds.getFeatures().size(0);
out.add(new Tuple2<>(numExamples, score * numExamples));
}

11
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark/src/main/java/org/deeplearning4j/spark/impl/graph/scoring/ScoreFlatMapFunctionCGMultiDataSet.java
View File

@ -35,7 +35,7 @@ import java.util.Iterator;
import java.util.List;
/** Function used to score a MultiDataSet using a given ComputationGraph */
public class ScoreFlatMapFunctionCGMultiDataSet implements FlatMapFunction<Iterator<MultiDataSet>, Tuple2<Integer, Double>> {
public class ScoreFlatMapFunctionCGMultiDataSet implements FlatMapFunction<Iterator<MultiDataSet>, Tuple2<Long, Double>> {
private static final Logger log = LoggerFactory.getLogger(ScoreFlatMapFunctionCGMultiDataSet.class);
private String json;
@ -50,9 +50,9 @@ public class ScoreFlatMapFunctionCGMultiDataSet implements FlatMapFunction<Itera
}
@Override
public Iterator<Tuple2<Integer, Double>> call(Iterator<MultiDataSet> dataSetIterator) throws Exception {
public Iterator<Tuple2<Long, Double>> call(Iterator<MultiDataSet> dataSetIterator) throws Exception {
if (!dataSetIterator.hasNext()) {
return Collections.singletonList(new Tuple2<>(0, 0.0)).iterator();
return Collections.singletonList(new Tuple2<>(0L, 0.0)).iterator();
}
MultiDataSetIterator iter = new IteratorMultiDataSetIterator(dataSetIterator, minibatchSize); //Does batching where appropriate
@ -66,13 +66,12 @@ public class ScoreFlatMapFunctionCGMultiDataSet implements FlatMapFunction<Itera
"Network did not have same number of parameters as the broadcast set parameters");
network.setParams(val);
List<Tuple2<Integer, Double>> out = new ArrayList<>();
List<Tuple2<Long, Double>> out = new ArrayList<>();
while (iter.hasNext()) {
MultiDataSet ds = iter.next();
double score = network.score(ds, false);
// FIXME: int cast
int numExamples = (int) ds.getFeatures(0).size(0);
long numExamples = ds.getFeatures(0).size(0);
out.add(new Tuple2<>(numExamples, score * numExamples));
}

1
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark/src/main/java/org/deeplearning4j/spark/impl/multilayer/scoring/FeedForwardWithKeyFunction.java
View File

@ -105,7 +105,6 @@ public class FeedForwardWithKeyFunction<K>
fMaskList.add(t2._2()._2());
keyList.add(t2._1());
// FIXME: int cast
origSizeList.add((int) t2._2()._1().size(0));
tupleCount++;
}

1
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark/src/main/java/org/deeplearning4j/spark/impl/multilayer/scoring/ScoreFlatMapFunction.java
View File

@ -64,7 +64,6 @@ public class ScoreFlatMapFunction implements FlatMapFunction<Iterator<DataSet>,
DataSet ds = iter.next();
double score = network.score(ds, false);
// FIXME: int cast
val numExamples = (int) ds.getFeatures().size(0);
out.add(new Tuple2<>(numExamples, score * numExamples));
}

4
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark/src/main/java/org/deeplearning4j/spark/impl/paramavg/ParameterAveragingTrainingWorker.java
View File

@ -247,10 +247,8 @@ public class ParameterAveragingTrainingWorker extends BaseTrainingWorker<Paramet
trainingHook.postUpdate(dataSet, graph);
}
}
// FIXME: int cast
if (configuration.isCollectTrainingStats())
stats.logFitEnd((int) dataSet.getFeatures(0).size(0));
stats.logFitEnd(dataSet.getFeatures(0).size(0));
Nd4j.getExecutioner().commit();

2
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark/src/main/java/org/deeplearning4j/spark/impl/paramavg/stats/ParameterAveragingTrainingWorkerStats.java
View File

@ -195,7 +195,7 @@ public class ParameterAveragingTrainingWorkerStats implements SparkTrainingStats
lastFitStartTime = timeSource.currentTimeMillis();
}
public void logFitEnd(int numExamples) {
public void logFitEnd(long numExamples) {
long now = timeSource.currentTimeMillis();
fitTimes.add(new ExampleCountEventStats(lastFitStartTime, now - lastFitStartTime, numExamples));
}

9
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark/src/main/java/org/deeplearning4j/spark/iterator/BaseDataSetIterator.java
View File

@ -19,6 +19,7 @@ package org.deeplearning4j.spark.iterator;
import org.nd4j.linalg.dataset.DataSet;
import org.nd4j.linalg.dataset.api.DataSetPreProcessor;
import org.nd4j.linalg.dataset.api.iterator.DataSetIterator;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import java.util.Collection;
import java.util.Iterator;
@ -31,8 +32,8 @@ public abstract class BaseDataSetIterator<T> implements DataSetIterator {
protected Collection<T> dataSetStreams;
protected DataSetPreProcessor preprocessor;
protected Iterator<T> iter;
protected int totalOutcomes = -1;
protected int inputColumns = -1;
protected long totalOutcomes = -1;
protected long inputColumns = -1;
protected int batch = -1;
protected DataSet preloadedDataSet;
protected int cursor = 0;
@ -112,7 +113,9 @@ public abstract class BaseDataSetIterator<T> implements DataSetIterator {
private void preloadDataSet() {
preloadedDataSet = load(iter.next());
// FIXME: int cast
if (preloadedDataSet.getLabels().size(1) > Integer.MAX_VALUE ||
preloadedDataSet.getFeatures().size(1) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
totalOutcomes = (int) preloadedDataSet.getLabels().size(1);
inputColumns = (int) preloadedDataSet.getFeatures().size(1);
}

1
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark/src/main/java/org/deeplearning4j/spark/iterator/PathSparkDataSetIterator.java
View File

@ -67,7 +67,6 @@ public class PathSparkDataSetIterator extends BaseDataSetIterator<String> {
ds = load(iter.next());
}
// FIXME: int cast
totalOutcomes = ds.getLabels() == null ? 0 : (int) ds.getLabels().size(1); //May be null for layerwise pretraining
inputColumns = (int) ds.getFeatures().size(1);
batch = ds.numExamples();

5
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark/src/main/java/org/deeplearning4j/spark/iterator/PortableDataStreamDataSetIterator.java
View File

@ -18,6 +18,7 @@ package org.deeplearning4j.spark.iterator;
import org.apache.spark.input.PortableDataStream;
import org.nd4j.linalg.dataset.DataSet;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import java.io.IOException;
import java.io.InputStream;
@ -53,7 +54,9 @@ public class PortableDataStreamDataSetIterator extends BaseDataSetIterator<Porta
ds = load(iter.next());
}
// FIXME: int cast
if (ds.getLabels().size(1) > Integer.MAX_VALUE ||
ds.getFeatures().size(1) > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
totalOutcomes = (int) ds.getLabels().size(1);
inputColumns = (int) ds.getFeatures().size(1);
batch = ds.numExamples();

4
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark/src/main/java/org/deeplearning4j/spark/stats/ExampleCountEventStats.java
View File

@ -26,9 +26,9 @@ import lombok.Getter;
public class ExampleCountEventStats extends BaseEventStats {
@Getter
private final int totalExampleCount;
private final long totalExampleCount;
public ExampleCountEventStats(long startTime, long durationMs, int totalExampleCount) {
public ExampleCountEventStats(long startTime, long durationMs, long totalExampleCount) {
super(startTime, durationMs);
this.totalExampleCount = totalExampleCount;
}

4
deeplearning4j/deeplearning4j-scaleout/spark/dl4j-spark/src/main/java/org/deeplearning4j/spark/util/MLLibUtil.java
View File

@ -31,6 +31,7 @@ import org.datavec.api.split.InputStreamInputSplit;
import org.datavec.api.writable.Writable;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.dataset.DataSet;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.util.FeatureUtil;
import scala.Tuple2;
@ -122,7 +123,8 @@ public class MLLibUtil {
if (!arr.isVector()) {
throw new IllegalArgumentException("passed in array must be a vector");
}
// FIXME: int cast
if (arr.length() > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
double[] ret = new double[(int) arr.length()];
for (int i = 0; i < arr.length(); i++) {
ret[i] = arr.getDouble(i);

10
deeplearning4j/deeplearning4j-ui-parent/deeplearning4j-ui/src/main/java/org/deeplearning4j/ui/weights/ConvolutionalIterationListener.java
View File

@ -33,6 +33,7 @@ import org.deeplearning4j.ui.api.UIServer;
import org.deeplearning4j.ui.storage.mapdb.MapDBStatsStorage;
import org.deeplearning4j.util.UIDProvider;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.io.ClassPathResource;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
@ -205,7 +206,8 @@ public class ConvolutionalIterationListener extends BaseTrainingListener {
if(layers[i].type() == Layer.Type.CONVOLUTIONAL){
INDArray output = activations.get(i+1); //Offset by 1 - activations list includes input
// FIXME: int cast
if (output.shape()[0] - 1 > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
int sampleDim = output.shape()[0] == 1 ? 0 : rnd.nextInt((int) output.shape()[0] - 1) + 1;
if (cnt == 0) {
INDArray inputs = layers[i].input();
@ -426,7 +428,8 @@ public class ConvolutionalIterationListener extends BaseTrainingListener {
val height = (numRows * (tShape[1] + border + padding_col)) + padding_col + zoomPadding + zoomWidth;
// FIXME: int cast
if (height > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
BufferedImage outputImage = new BufferedImage(maxWidth, (int) height, BufferedImage.TYPE_BYTE_GRAY);
Graphics2D graphics2D = outputImage.createGraphics();
@ -571,7 +574,8 @@ public class ConvolutionalIterationListener extends BaseTrainingListener {
*/
graphics2D.setPaint(borderColor);
// FIXME: int cast
if (tad2D.shape()[0] > Integer.MAX_VALUE || tad2D.shape()[1] > Integer.MAX_VALUE)
throw new ND4JArraySizeException();
graphics2D.drawRect(columnOffset, rowOffset, (int) tad2D.shape()[0], (int) tad2D.shape()[1]);

5
deeplearning4j/deeplearning4j-zoo/src/main/java/org/deeplearning4j/zoo/util/BaseLabels.java
View File

@ -90,9 +90,8 @@ public abstract class BaseLabels implements Labels {
Preconditions.checkState(predictions.size(1) == labels.size(), "Invalid input array:" +
" expected array with size(1) equal to numLabels (%s), got array with shape %s", labels.size(), predictions.shape());
// FIXME: int cast
int rows = (int) predictions.size(0);
int cols = (int) predictions.size(1);
long rows = predictions.size(0);
long cols = predictions.size(1);
if (predictions.isColumnVectorOrScalar()) {
predictions = predictions.ravel();
rows = (int) predictions.size(0);

1
deeplearning4j/dl4j-integration-tests/src/test/java/org/deeplearning4j/integration/TestUtils.java
View File

@ -116,7 +116,6 @@ public class TestUtils {
public static INDArray randomOneHot(long examples, long nOut, Random rng){
INDArray arr = Nd4j.create(examples, nOut);
for( int i=0; i<examples; i++ ){
// FIXME: int cast
arr.putScalar(i, rng.nextInt((int) nOut), 1.0);
}
return arr;

Some files were not shown because too many files have changed in this diff Show More