Java Code Examples for org.nd4j.autodiff.samediff.SameDiff#setTrainingConfig()

The following examples show how to use org.nd4j.autodiff.samediff.SameDiff#setTrainingConfig() . You can vote up the ones you like or vote down the ones you don't like, and go to the original project or source file by following the links above each example. You may check out the related API usage on the sidebar.
Example 1
Source File: CheckpointListenerTest.java    From deeplearning4j with Apache License 2.0 6 votes vote down vote up 
public static SameDiff getModel(){
    Nd4j.getRandom().setSeed(12345);
    SameDiff sd = SameDiff.create();
    SDVariable in = sd.placeHolder("in", DataType.FLOAT, -1, 4);
    SDVariable label = sd.placeHolder("label", DataType.FLOAT, -1, 3);
    SDVariable w = sd.var("W", Nd4j.rand(DataType.FLOAT, 4, 3));
    SDVariable b = sd.var("b", DataType.FLOAT, 3);

    SDVariable mmul = in.mmul(w).add(b);
    SDVariable softmax = sd.nn().softmax(mmul);
    SDVariable loss = sd.loss().logLoss("loss", label, softmax);

    sd.setTrainingConfig(TrainingConfig.builder()
            .dataSetFeatureMapping("in")
            .dataSetLabelMapping("label")
            .updater(new Adam(1e-2))
            .weightDecay(1e-2, true)
            .build());

    return sd;
}
Example 2
Source File: UIListenerTest.java    From deeplearning4j with Apache License 2.0 6 votes vote down vote up 
private static SameDiff getSimpleNet(){
    Nd4j.getRandom().setSeed(12345);
    SameDiff sd = SameDiff.create();
    SDVariable in = sd.placeHolder("in", DataType.FLOAT, -1, 4);
    SDVariable label = sd.placeHolder("label", DataType.FLOAT, -1, 3);
    SDVariable w = sd.var("W", Nd4j.rand(DataType.FLOAT, 4, 3));
    SDVariable b = sd.var("b", DataType.FLOAT, 1, 3);
    SDVariable mmul = in.mmul(w).add(b);
    SDVariable softmax = sd.nn.softmax("softmax", mmul);
    SDVariable loss = sd.loss().logLoss("loss", label, softmax);

    sd.setTrainingConfig(TrainingConfig.builder()
            .dataSetFeatureMapping("in")
            .dataSetLabelMapping("label")
            .updater(new Adam(1e-1))
            .weightDecay(1e-3, true)
            .build());
    return sd;
}
Example 3
Source File: ListenerTest.java    From deeplearning4j with Apache License 2.0 5 votes vote down vote up 
@Test
public void testCustomListener() {
    SameDiff sd = SameDiff.create();
    SDVariable in = sd.placeHolder("input", DataType.FLOAT, -1, 4);
    SDVariable label = sd.placeHolder("label", DataType.FLOAT, -1, 3);
    SDVariable w = sd.var("w", Nd4j.rand(DataType.FLOAT, 4, 3));
    SDVariable b = sd.var("b", Nd4j.rand(DataType.FLOAT, 3));
    SDVariable z = sd.nn().linear("z", in, w, b);
    SDVariable out = sd.nn().softmax("out", z, 1);
    SDVariable loss = sd.loss().softmaxCrossEntropy("loss", label, out, null);

    //Create and set the training configuration
    double learningRate = 1e-3;
    TrainingConfig config = new TrainingConfig.Builder()
            .l2(1e-4)                               //L2 regularization
            .updater(new Adam(learningRate))        //Adam optimizer with specified learning rate
            .dataSetFeatureMapping("input")         //DataSet features array should be associated with variable "input"
            .dataSetLabelMapping("label")           //DataSet label array should be associated with variable "label
            .addEvaluations(false,"out",0,new Evaluation())
            .build();
    sd.setTrainingConfig(config);

    CustomListener listener = new CustomListener();
    Map<String,INDArray> m = sd.output()
            .data(new IrisDataSetIterator(150, 150))
            .output("out")
            .listeners(listener)
            .exec();

    assertEquals(1, m.size());
    assertTrue(m.containsKey("out"));
    assertNotNull(listener.z);
    assertNotNull(listener.out);

}
Example 4
Source File: ExecDebuggingListenerTest.java    From deeplearning4j with Apache License 2.0 5 votes vote down vote up 
@Test
    public void testExecDebugListener(){

        SameDiff sd = SameDiff.create();
        SDVariable in = sd.placeHolder("in", DataType.FLOAT, -1, 3);
        SDVariable label = sd.placeHolder("label", DataType.FLOAT, 1, 2);
        SDVariable w = sd.var("w", Nd4j.rand(DataType.FLOAT, 3, 2));
        SDVariable b = sd.var("b", Nd4j.rand(DataType.FLOAT, 1, 2));
        SDVariable sm = sd.nn.softmax("softmax", in.mmul(w).add(b));
        SDVariable loss = sd.loss.logLoss("loss", label, sm);

        INDArray i = Nd4j.rand(DataType.FLOAT, 1, 3);
        INDArray l = Nd4j.rand(DataType.FLOAT, 1, 2);

        sd.setTrainingConfig(TrainingConfig.builder()
                .dataSetFeatureMapping("in")
                .dataSetLabelMapping("label")
                .updater(new Adam(0.001))
                .build());

        for(ExecDebuggingListener.PrintMode pm : ExecDebuggingListener.PrintMode.values()){
            sd.setListeners(new ExecDebuggingListener(pm, -1, true));
//            sd.output(m, "softmax");
            sd.fit(new DataSet(i, l));

            System.out.println("\n\n\n");
        }

    }
Example 5
Source File: UIListenerTest.java    From deeplearning4j with Apache License 2.0 5 votes vote down vote up 
@Test
public void testUIListenerBasic() throws Exception {
    Nd4j.getRandom().setSeed(12345);

    IrisDataSetIterator iter = new IrisDataSetIterator(150, 150);

    SameDiff sd = getSimpleNet();

    File dir = testDir.newFolder();
    File f = new File(dir, "logFile.bin");
    UIListener l = UIListener.builder(f)
            .plotLosses(1)
            .trainEvaluationMetrics("softmax", 0, Evaluation.Metric.ACCURACY, Evaluation.Metric.F1)
            .updateRatios(1)
            .build();

    sd.setListeners(l);

    sd.setTrainingConfig(TrainingConfig.builder()
            .dataSetFeatureMapping("in")
            .dataSetLabelMapping("label")
            .updater(new Adam(1e-1))
            .weightDecay(1e-3, true)
            .build());

    sd.fit(iter, 20);

    //Test inference after training with UI Listener still around
    Map<String, INDArray> m = new HashMap<>();
    iter.reset();
    m.put("in", iter.next().getFeatures());
    INDArray out = sd.outputSingle(m, "softmax");
    assertNotNull(out);
    assertArrayEquals(new long[]{150, 3}, out.shape());
}
Example 6
Source File: SameDiffRNNTestCases.java    From deeplearning4j with Apache License 2.0 4 votes vote down vote up 
@Override
        public Object getConfiguration() throws Exception {
            Nd4j.getRandom().setSeed(12345);


            int miniBatchSize = 10;
            int numLabelClasses = 6;
            int nIn = 60;
            int numUnits = 7;
            int timeSteps = 3;


            SameDiff sd = SameDiff.create();

            SDVariable in = sd.placeHolder("in", DataType.FLOAT, miniBatchSize, timeSteps, nIn);
            SDVariable label = sd.placeHolder("label", DataType.FLOAT, miniBatchSize, numLabelClasses);


            SDVariable cLast = sd.var("cLast", Nd4j.zeros(DataType.FLOAT, miniBatchSize, numUnits));
            SDVariable yLast = sd.var("yLast", Nd4j.zeros(DataType.FLOAT, miniBatchSize, numUnits));

            LSTMLayerConfig c = LSTMLayerConfig.builder()
                    .lstmdataformat(LSTMDataFormat.NTS)
                    .directionMode(LSTMDirectionMode.FWD)
                    .gateAct(LSTMActivations.SIGMOID)
                    .cellAct(LSTMActivations.TANH)
                    .outAct(LSTMActivations.TANH)
                    .retFullSequence(true)
                    .retLastC(true)
                    .retLastH(true)
                    .build();

            LSTMLayerOutputs outputs = new LSTMLayerOutputs(sd.rnn.lstmLayer(
                    in, cLast, yLast, null,
                    LSTMLayerWeights.builder()
                            .weights(sd.var("weights", Nd4j.rand(DataType.FLOAT, nIn, 4 * numUnits)))
                            .rWeights(sd.var("rWeights", Nd4j.rand(DataType.FLOAT, numUnits, 4 * numUnits)))
                            .peepholeWeights(sd.var("inputPeepholeWeights", Nd4j.rand(DataType.FLOAT, 3 * numUnits)))
                            .bias(sd.var("bias", Nd4j.rand(DataType.FLOAT, 4 * numUnits)))
                            .build(),
                    c), c);


//           Behaviour with default settings: 3d (time series) input with shape
//          [miniBatchSize, vectorSize, timeSeriesLength] -> 2d output [miniBatchSize, vectorSize]
            SDVariable layer0 = outputs.getOutput();

            SDVariable layer1 = layer0.mean(1);

            SDVariable w1 = sd.var("w1", Nd4j.rand(DataType.FLOAT, numUnits, numLabelClasses));
            SDVariable b1 = sd.var("b1", Nd4j.rand(DataType.FLOAT, numLabelClasses));


            SDVariable out = sd.nn.softmax("out", layer1.mmul(w1).add(b1));
            SDVariable loss = sd.loss.logLoss("loss", label, out);

            //Also set the training configuration:
            sd.setTrainingConfig(TrainingConfig.builder()
                    .updater(new Adam(5e-2))
                    .l1(1e-3).l2(1e-3)
                    .dataSetFeatureMapping("in")            //features[0] -> "in" placeholder
                    .dataSetLabelMapping("label")           //labels[0]   -> "label" placeholder
                    .build());

            return sd;

        }