Examples with SDVariable org.nd4j.autodiff.samediff.SDVariable used on opensource projects

Example 71 with SDVariable

use of org.nd4j.autodiff.samediff.SDVariable in project nd4j by deeplearning4j.

the class Norm1 method doDiff.

@Override
public List<SDVariable> doDiff(List<SDVariable> i_v1) {
    // d l1Norm(in)/dx = signum(x)
    SDVariable signum = sameDiff.sign(arg());
    // Note that we need to expand the dimensions of the gradient - auto-broadcast won't work for all cases.
    // TODO shape may not always be defined?
    int origRank = Shape.rankFromShape(arg().getShape());
    SDVariable bcGrad = sameDiff.f().reductionBroadcastableWithOrigShape(origRank, dimensions, i_v1.get(0));
    return Arrays.asList(signum.mul(bcGrad));
}

Example 72 with SDVariable

use of org.nd4j.autodiff.samediff.SDVariable in project nd4j by deeplearning4j.

the class Norm2 method doDiff.

@Override
public List<SDVariable> doDiff(List<SDVariable> i_v1) {
    // d norm2(in)/dx = x / norm2(in)
    SDVariable norm2 = outputVariables()[0];
    // TODO shape may not always be defined?
    int origRank = Shape.rankFromShape(arg().getShape());
    SDVariable broadcastableNorm2 = f().reductionBroadcastableWithOrigShape(origRank, dimensions, norm2);
    SDVariable broadcastableGradOut = f().reductionBroadcastableWithOrigShape(origRank, dimensions, i_v1.get(0));
    SDVariable ret = arg().div(broadcastableNorm2).mul(broadcastableGradOut);
    return Arrays.asList(ret);
}

Example 73 with SDVariable

use of org.nd4j.autodiff.samediff.SDVariable in project nd4j by deeplearning4j.

the class GradCheckTransforms method testDepthToSpace.

@Test
public void testDepthToSpace() {
    Nd4j.getRandom().setSeed(1337);
    int miniBatch = 128;
    int blockSize = 4;
    String dataFormat = "NHWC";
    int isNHWC = dataFormat.equals("NHWC") ? 1 : 0;
    int[] inputShape = new int[] { miniBatch, 2, 2, blockSize * blockSize };
    INDArray input = Nd4j.randn(inputShape);
    SameDiff sd = SameDiff.create();
    SDVariable sdInput = sd.var("in", inputShape);
    INDArray expOut = Nd4j.create(miniBatch, 2 * blockSize, 2 * blockSize, 1);
    DynamicCustomOp op = DynamicCustomOp.builder("depth_to_space").addInputs(input).addIntegerArguments(blockSize, isNHWC).addOutputs(expOut).build();
    Nd4j.getExecutioner().exec(op);
    sd.associateArrayWithVariable(input, sdInput);
    SDVariable t = sd.depthToSpace(sdInput, blockSize, dataFormat);
    SDVariable loss = sd.mean("loss", t);
    sd.exec();
    INDArray out = t.getArr();
    if (!expOut.equals(out)) {
        log.info("depth to space failed on forward");
    }
    try {
        GradCheckUtil.checkGradients(sd);
    } catch (Exception e) {
        e.printStackTrace();
    }
}

Example 74 with SDVariable

use of org.nd4j.autodiff.samediff.SDVariable in project nd4j by deeplearning4j.

the class GradCheckTransforms method testCross.

@Test
public void testCross() {
    INDArray a = Nd4j.create(new float[] { 4, 2, 1 }, new int[] { 1, 3 });
    INDArray b = Nd4j.create(new float[] { 1, 3, 4 }, new int[] { 1, 3 });
    INDArray expOut = Nd4j.create(1, 3);
    DynamicCustomOp op = DynamicCustomOp.builder("cross").addInputs(a, b).addOutputs(expOut).build();
    Nd4j.getExecutioner().exec(op);
    SameDiff sd = SameDiff.create();
    SDVariable sdA = sd.var("a", expOut.shape());
    SDVariable sdB = sd.var("b", expOut.shape());
    sd.associateArrayWithVariable(a, sdA);
    sd.associateArrayWithVariable(b, sdB);
    SDVariable t = sd.cross(sdA, sdB);
    SDVariable loss = sd.mean("loss", t);
    sd.exec();
    INDArray out = t.getArr();
    if (!expOut.equals(out)) {
        log.info("batch to space failed on forward");
    }
    try {
        GradCheckUtil.checkGradients(sd);
    } catch (Exception e) {
        e.printStackTrace();
    }
}

Example 75 with SDVariable

use of org.nd4j.autodiff.samediff.SDVariable in project nd4j by deeplearning4j.

the class GradCheckTransforms method testSpaceToBatch.

@Test
public void testSpaceToBatch() {
    Nd4j.getRandom().setSeed(7331);
    int miniBatch = 4;
    int[] inputShape = new int[] { 1, 2, 2, 1 };
    int M = 2;
    int[] blockShape = new int[] { M, 1 };
    int[] paddingShape = new int[] { M, 2 };
    INDArray input = Nd4j.randn(inputShape);
    INDArray blocks = Nd4j.create(new float[] { 2, 2 }, blockShape);
    INDArray padding = Nd4j.create(new float[] { 0, 0, 0, 0 }, paddingShape);
    SameDiff sd = SameDiff.create();
    SDVariable sdInput = sd.var("in", inputShape);
    INDArray expOut = Nd4j.create(miniBatch, 1, 1, 1);
    DynamicCustomOp op = DynamicCustomOp.builder("space_to_batch").addInputs(input, blocks, padding).addOutputs(expOut).build();
    Nd4j.getExecutioner().exec(op);
    sd.associateArrayWithVariable(input, sdInput);
    SDVariable t = sd.spaceToBatch(sdInput, new int[] { 2, 2 }, new int[][] { { 0, 0 }, { 0, 0 } });
    SDVariable loss = sd.mean("loss", t);
    sd.exec();
    INDArray out = t.getArr();
    if (!expOut.equals(out)) {
        log.info("space to batch failed on forward");
    }
    try {
        GradCheckUtil.checkGradients(sd);
    } catch (Exception e) {
        e.printStackTrace();
    }
}