// TODO remove/move, technically this is testing Nd4j functionality
@Test
public void testCreateFeatureMapMethod() {
Layer layer = getContainedConfig();
INDArray input = getContainedData();
int inputWidth = input.shape()[0];
int featureMapWidth =
(inputWidth + layer.conf().getPadding()[0] * 2 - layer.conf().getKernelSize()[0])
/ layer.conf().getStride()[0]
+ 1;
INDArray expectedOutput =
Nd4j.create(
new double[] {
1, 1, 1, 1, 3, 3, 3, 3, 1, 1, 1, 1, 3, 3, 3, 3, 1, 1, 1, 1, 3, 3, 3,
3, 1, 1, 1, 1, 3, 3, 3, 3, 2, 2, 2, 2, 4, 4, 4, 4, 2, 2, 2, 2, 4, 4,
4, 4, 2, 2, 2, 2, 4, 4, 4, 4, 2, 2, 2, 2, 4, 4, 4, 4
},
new int[] {1, 1, 2, 2, 4, 4});
layer.setInput(input);
org.deeplearning4j.nn.layers.convolution.ConvolutionLayer layer2 =
(org.deeplearning4j.nn.layers.convolution.ConvolutionLayer) layer;
INDArray featureMaps = layer2.createFeatureMapColumn();
assertEquals(featureMapWidth, featureMaps.shape()[4]);
assertEquals(expectedOutput.shape(), featureMaps.shape());
assertEquals(expectedOutput, featureMaps);
}
// note precision is off on this test but the numbers are close
// investigation in a future release should determine how to resolve
@Test
public void testCalculateDeltaContained() {
Layer layer = getContainedConfig();
INDArray input = getContainedData();
INDArray col = getContainedCol();
INDArray epsilon = Nd4j.ones(1, 2, 4, 4);
INDArray expectedOutput =
Nd4j.create(
new double[] {
0.02036651, 0.02036651, 0.02036651, 0.02036651, 0.00039383,
0.00039383, 0.00039383, 0.00039383, 0.02036651, 0.02036651,
0.02036651, 0.02036651, 0.00039383, 0.00039383, 0.00039383,
0.00039383, 0.02036651, 0.02036651, 0.02036651, 0.02036651,
0.00039383, 0.00039383, 0.00039383, 0.00039383, 0.02036651,
0.02036651, 0.02036651, 0.02036651, 0.00039383, 0.00039383,
0.00039383, 0.00039383
},
new int[] {1, 2, 4, 4});
layer.setInput(input);
org.deeplearning4j.nn.layers.convolution.ConvolutionLayer layer2 =
(org.deeplearning4j.nn.layers.convolution.ConvolutionLayer) layer;
layer2.setCol(col);
INDArray delta = layer2.calculateDelta(epsilon);
assertArrayEquals(expectedOutput.shape(), delta.shape());
assertEquals(expectedOutput, delta);
}
// note precision is off on this test but the numbers are close
// investigation in a future release should determine how to resolve
@Test
public void testBackpropResultsContained() {
Layer layer = getContainedConfig();
INDArray input = getContainedData();
INDArray col = getContainedCol();
INDArray epsilon = Nd4j.ones(1, 2, 4, 4);
INDArray expectedBiasGradient =
Nd4j.create(new double[] {0.16608272, 0.16608272}, new int[] {1, 2});
INDArray expectedWeightGradient =
Nd4j.create(
new double[] {
0.17238397,
0.17238397,
0.33846668,
0.33846668,
0.17238397,
0.17238397,
0.33846668,
0.33846668
},
new int[] {2, 1, 2, 2});
INDArray expectedEpsilon =
Nd4j.create(
new double[] {
0.00039383, 0.00039383, 0.00039383, 0.00039383, 0.00039383,
0.00039383, 0., 0., 0.00039383, 0.00039383,
0.00039383, 0.00039383, 0.00039383, 0.00039383, 0.,
0., 0.02036651, 0.02036651, 0.02036651, 0.02036651,
0.02036651, 0.02036651, 0., 0., 0.02036651,
0.02036651, 0.02036651, 0.02036651, 0.02036651, 0.02036651,
0., 0., 0.00039383, 0.00039383, 0.00039383,
0.00039383, 0.00039383, 0.00039383, 0., 0.,
0.00039383, 0.00039383, 0.00039383, 0.00039383, 0.00039383,
0.00039383, 0., 0., 0., 0.,
0., 0., 0., 0., 0.,
0., 0., 0., 0., 0.,
0., 0., 0., 0.
},
new int[] {1, 1, 8, 8});
layer.setInput(input);
org.deeplearning4j.nn.layers.convolution.ConvolutionLayer layer2 =
(org.deeplearning4j.nn.layers.convolution.ConvolutionLayer) layer;
layer2.setCol(col);
Pair<Gradient, INDArray> pair = layer2.backpropGradient(epsilon);
assertArrayEquals(expectedEpsilon.shape(), pair.getSecond().shape());
assertArrayEquals(expectedWeightGradient.shape(), pair.getFirst().getGradientFor("W").shape());
assertArrayEquals(expectedBiasGradient.shape(), pair.getFirst().getGradientFor("b").shape());
assertEquals(expectedEpsilon, pair.getSecond());
assertEquals(expectedWeightGradient, pair.getFirst().getGradientFor("W"));
assertEquals(expectedBiasGradient, pair.getFirst().getGradientFor("b"));
}
@Test
public void testPreOutputMethodContained() {
Layer layer = getContainedConfig();
INDArray col = getContainedCol();
INDArray expectedOutput =
Nd4j.create(
new double[] {
4., 4., 4., 4., 8., 8., 8., 8., 4., 4., 4., 4., 8., 8., 8., 8., 4., 4., 4., 4., 8.,
8., 8., 8., 4., 4., 4., 4., 8., 8., 8., 8
},
new int[] {1, 2, 4, 4});
org.deeplearning4j.nn.layers.convolution.ConvolutionLayer layer2 =
(org.deeplearning4j.nn.layers.convolution.ConvolutionLayer) layer;
layer2.setCol(col);
INDArray activation = layer2.preOutput(true);
assertArrayEquals(expectedOutput.shape(), activation.shape());
assertEquals(expectedOutput, activation);
}
@Test
public void testCalculateDelta() {
Layer layer = getContainedConfig();
INDArray col = getContainedCol();
INDArray expectedOutput =
Nd4j.create(
new double[] {
-12., -12., -12., -12., -56., -56., -56., -56., -12., -12., -12.,
-12., -56., -56., -56., -56., -12., -12., -12., -12., -56., -56.,
-56., -56., -12., -12., -12., -12., -56., -56., -56., -56.
},
new int[] {1, 2, 4, 4});
org.deeplearning4j.nn.layers.convolution.ConvolutionLayer layer2 =
(org.deeplearning4j.nn.layers.convolution.ConvolutionLayer) layer;
layer2.setCol(col);
INDArray delta = layer2.calculateDelta(epsilon);
assertEquals(expectedOutput.shape(), delta.shape());
assertEquals(expectedOutput, delta);
}
@Test
public void testBackpropResults() {
Layer layer = getContainedConfig();
INDArray col = getContainedCol();
INDArray expectedWeightGradient =
Nd4j.create(
new double[] {-1440., -1440., -1984., -1984., -1440., -1440., -1984., -1984.},
new int[] {2, 1, 2, 2});
INDArray expectedBiasGradient =
Nd4j.create(
new double[] {-544., -544.},
new int[] {
2,
});
INDArray expectedEpsilon =
Nd4j.create(
new double[] {
-12., -12., -12., -12., -12., -12., -12., -12., -12., -12., -12.,
-12., -12., -12., -12., -12., -56., -56., -56., -56., -56., -56.,
-56., -56., -56., -56., -56., -56., -56., -56., -56., -56., -12.,
-12., -12., -12., -12., -12., -12., -12., -12., -12., -12., -12.,
-12., -12., -12., -12., -56., -56., -56., -56., -56., -56., -56.,
-56., -56., -56., -56., -56., -56., -56., -56., -56.
},
new int[] {1, 1, 8, 8});
org.deeplearning4j.nn.layers.convolution.ConvolutionLayer layer2 =
(org.deeplearning4j.nn.layers.convolution.ConvolutionLayer) layer;
layer2.setCol(col);
Pair<Gradient, INDArray> pair = layer2.backpropGradient(epsilon);
assertEquals(expectedEpsilon.shape(), pair.getSecond().shape());
assertEquals(expectedWeightGradient.shape(), pair.getFirst().getGradientFor("W").shape());
assertEquals(expectedBiasGradient.shape(), pair.getFirst().getGradientFor("b").shape());
assertEquals(expectedEpsilon, pair.getSecond());
assertEquals(expectedWeightGradient, pair.getFirst().getGradientFor("W"));
assertEquals(expectedBiasGradient, pair.getFirst().getGradientFor("b"));
}
public static void testAccuracy() {
double[][][][] data = {
{
{
{1.0, 2.0, 3.0},
{4.0, 5.0, 6.0},
{7.0, 8.0, 9.0}
}
}
};
double[] flat = ArrayUtil.flattenDoubleArray(data);
int[] shape = {1, 1, 3, 3};
INDArray input = Nd4j.create(flat, shape, 'c');
TestCase testCase = new TestCase(1, 1, 2, 2, 1, 1, 0, 0, 3, 3);
ConvolutionLayer convolutionLayerBuilder =
new ConvolutionLayer.Builder(testCase.kW, testCase.kH)
.nIn(testCase.nInputPlane)
.stride(testCase.dW, testCase.dH)
.padding(testCase.padW, testCase.padH)
.nOut(testCase.nOutputPlane)
.build();
MultiLayerConfiguration.Builder builder =
new NeuralNetConfiguration.Builder().list().layer(0, convolutionLayerBuilder);
MultiLayerConfiguration conf = builder.build();
MultiLayerNetwork model = new MultiLayerNetwork(conf);
model.init();
model.setInput(input);
model.getLayer(0).setInput(input);
org.deeplearning4j.nn.layers.convolution.ConvolutionLayer convolutionLayer =
(org.deeplearning4j.nn.layers.convolution.ConvolutionLayer) model.getLayer(0);
System.out.println(convolutionLayer.params());
System.out.println(convolutionLayer.preOutput(false));
}
public static void testForward() {
for (TestCase testCase : allTestCases) {
try (BufferedWriter writer =
new BufferedWriter(new FileWriter(new File("dl4jPerformance.csv"), true))) {
ConvolutionLayer convolutionLayerBuilder =
new ConvolutionLayer.Builder(testCase.kW, testCase.kH)
.nIn(testCase.nInputPlane)
.stride(testCase.dW, testCase.dH)
.padding(testCase.padW, testCase.padH)
.nOut(testCase.nOutputPlane)
.build();
MultiLayerConfiguration.Builder builder =
new NeuralNetConfiguration.Builder().list().layer(0, convolutionLayerBuilder);
MultiLayerConfiguration conf = builder.build();
MultiLayerNetwork model = new MultiLayerNetwork(conf);
model.init();
INDArray input =
Nd4j.rand(
seed, batchSize, testCase.nInputPlane, testCase.inputWidth, testCase.inputHeight);
model.setInput(input);
model.getLayer(0).setInput(input);
org.deeplearning4j.nn.layers.convolution.ConvolutionLayer convolutionLayer =
(org.deeplearning4j.nn.layers.convolution.ConvolutionLayer) model.getLayer(0);
double start = System.nanoTime();
for (int i = 0; i < forwardIterations; i++) {
convolutionLayer.preOutput(false);
}
double end = System.nanoTime();
double timeMillis = (end - start) / 1e6 / forwardIterations;
writer.write(
"Convolution("
+ testCase.nInputPlane
+ " "
+ testCase.nOutputPlane
+ " "
+ testCase.kW
+ " "
+ testCase.kH
+ " "
+ testCase.dW
+ " "
+ testCase.dH
+ " "
+ testCase.padW
+ " "
+ testCase.padH
+ " "
+ testCase.inputWidth
+ " "
+ testCase.inputHeight
+ ") "
+ " forward, "
+ timeMillis
+ "\n");
} catch (Exception ex) {
ex.printStackTrace();
}
}
}
