@Test
public void testCNNInputSetupMNIST() throws Exception {
INDArray input = getMnistData();
Layer layer = getMNISTConfig();
layer.activate(input);
assertEquals(input, layer.input());
assertArrayEquals(input.shape(), layer.input().shape());
}
@Test
public void testFeatureMapShape() throws Exception {
INDArray input = getMnistData();
Layer layer = getCNNConfig(nChannelsIn, depth, kernelSize, stride, padding);
INDArray convActivations = layer.activate(input);
assertEquals(featureMapWidth, convActivations.size(2));
assertEquals(depth, convActivations.size(0));
}
@Test
public void testCNNInputSetup() throws Exception {
INDArray input = getMnistData();
int[] stride = new int[] {3, 3};
int[] padding = new int[] {1, 1};
Layer layer = getCNNConfig(nChannelsIn, depth, kernelSize, stride, padding);
layer.activate(input);
assertEquals(input, layer.input());
assertEquals(input.shape(), layer.input().shape());
}
@Test
public void testFeatureMapShapeMNIST() throws Exception {
int inputWidth = 28;
int[] stride = new int[] {2, 2};
int[] padding = new int[] {0, 0};
int[] kernelSize = new int[] {9, 9};
int nChannelsIn = 1;
int depth = 20;
int featureMapWidth = (inputWidth + padding[1] * 2 - kernelSize[1]) / stride[1] + 1;
INDArray input = getMnistData();
Layer layer = getCNNConfig(nChannelsIn, depth, kernelSize, stride, padding);
INDArray convActivations = layer.activate(input);
assertEquals(featureMapWidth, convActivations.size(2));
assertEquals(depth, convActivations.size(1));
}
@Test
public void testActivateResults() {
Layer layer = getContainedConfig();
INDArray input = getContainedData();
INDArray expectedOutput =
Nd4j.create(
new double[] {
0.98201379, 0.98201379, 0.98201379, 0.98201379, 0.99966465,
0.99966465, 0.99966465, 0.99966465, 0.98201379, 0.98201379,
0.98201379, 0.98201379, 0.99966465, 0.99966465, 0.99966465,
0.99966465, 0.98201379, 0.98201379, 0.98201379, 0.98201379,
0.99966465, 0.99966465, 0.99966465, 0.99966465, 0.98201379,
0.98201379, 0.98201379, 0.98201379, 0.99966465, 0.99966465,
0.99966465, 0.99966465
},
new int[] {1, 2, 4, 4});
INDArray convActivations = layer.activate(input);
assertEquals(expectedOutput, convActivations);
assertEquals(expectedOutput.shape(), convActivations.shape());
}
public static void testBackward() {
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);
model.feedForward();
org.deeplearning4j.nn.api.Layer convolutionLayer = model.getLayer(0);
INDArray output = convolutionLayer.activate();
INDArray epsilon =
Nd4j.rand(seed, output.size(0), output.size(1), output.size(2), output.size(3));
Method initGradientView = model.getClass().getDeclaredMethod("initGradientsView");
initGradientView.setAccessible(true);
initGradientView.invoke(model);
double start = System.nanoTime();
for (int i = 0; i < backwardIterations; i++) {
convolutionLayer.backpropGradient(epsilon);
}
double end = System.nanoTime();
double timeMillis = (end - start) / 1e6 / backwardIterations;
writer.write(
"Convolution("
+ testCase.nInputPlane
+ " "
+ testCase.nOutputPlane
+ " "
+ testCase.kW
+ " "
+ testCase.kH
+ " "
+ testCase.dW
+ " "
+ testCase.dH
+ " "
+ testCase.padW
+ " "
+ testCase.padH
+ " "
+ testCase.inputWidth
+ " "
+ testCase.inputHeight
+ ") "
+ " backward, "
+ timeMillis
+ "\n");
} catch (Exception ex) {
ex.printStackTrace();
}
}
}
