Hi, ich versuche einen Kursverlauf vorherzusagen... Die Vorhersage des Kurs springt jedoch "mit einem Peak" nach oben. Weiß jemand wieso?
Hier ist "der Knick" zu sehen:
Hier ist "der Knick" zu sehen:
Java:
private void predict(List<Double> tprices1, String symbol) {
System.out.println("Create data");
final int n1 = tprices1.size(); // input data length, 168 hours, one week
final int n2 = 64; // batch length
final int n3 = 24; // sample length, 24 hours, one day
final int n4 = 48; // prediction length, 48 hours, two days
double min1 = Double.MAX_VALUE;
double max1 = Double.MIN_VALUE;
for (Double double1 : tprices1) {
if (double1 < min1) {
min1 = double1;
}
if (double1 > max1) {
max1 = double1;
}
}
for (int i = 0; i < n4; i++) {
tprices1.add(tprices1.get(n1 - 1));
}
List<DataSet> dataSets0 = new ArrayList<>();
for (int k = 0; k < n1 + n4 - n2 - n3 - 1; k++) {
INDArray input = Nd4j.create(new int[] { n2, 1, n3 }, 'f');
INDArray label = Nd4j.create(new int[] { n2, 1, n3 }, 'f');
for (int i = 0; i < n2; i++) {
for (int j = 0; j < n3; j++) {
input.putScalar(new int[] { i, 0, j }, (tprices1.get(k + i + j) - min1) / (max1 - min1));
label.putScalar(new int[] { i, 0, j }, (tprices1.get(k + i + j + 1) - min1) / (max1 - min1));
}
}
dataSets0.add(new DataSet(input, label));
}
List<DataSet> dataSets1 = dataSets0.subList(0, n1 - n2 - n3);
List<DataSet> dataSets2 = dataSets0.subList(n1 - n2 - n3 - 1, dataSets0.size());
System.out.println(dataSets1.size());
System.out.println(dataSets2.size());
System.out.println("Build lstm networks");
final int nIn = 1;
final int nOut = 1;
final int seed = 345;
final int iterations = 1;
final double learningRate = 0.05;
final int lstmLayer1Size = 256;
final int lstmLayer2Size = 256;
final int denseLayerSize = 32;
final double dropoutRatio = 0.2;
final int truncatedBPTTLength = n3;
MultiLayerConfiguration conf = new NeuralNetConfiguration.Builder().seed(seed).iterations(iterations).learningRate(learningRate).optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT).weightInit(WeightInit.XAVIER)
.updater(Updater.RMSPROP).regularization(true).l2(1e-4).list().layer(0, new GravesLSTM.Builder().nIn(nIn).nOut(lstmLayer1Size).activation(Activation.TANH).gateActivationFunction(Activation.HARDSIGMOID).dropOut(dropoutRatio).build())
.layer(1, new GravesLSTM.Builder().nIn(lstmLayer1Size).nOut(lstmLayer2Size).activation(Activation.TANH).gateActivationFunction(Activation.HARDSIGMOID).dropOut(dropoutRatio).build())
.layer(2, new DenseLayer.Builder().nIn(lstmLayer2Size).nOut(denseLayerSize).activation(Activation.RELU).build())
.layer(3, new RnnOutputLayer.Builder().nIn(denseLayerSize).nOut(nOut).activation(Activation.IDENTITY).lossFunction(LossFunctions.LossFunction.MSE).build()).backpropType(BackpropType.TruncatedBPTT)
.tBPTTForwardLength(truncatedBPTTLength).tBPTTBackwardLength(truncatedBPTTLength).pretrain(false).backprop(true).build();
MultiLayerNetwork net = new MultiLayerNetwork(conf);
net.init();
net.setListeners(new ScoreIterationListener(10));
System.out.println("train...");
for (int epoch = 0; epoch < 1; epoch++) { // only one epoch
System.out.println("Epoch: " + epoch);
for (int i = 0; i < dataSets1.size(); i++) { // skip the first 0 data
net.fit(dataSets1.get(i)); // fit model using mini-batch data
}
net.rnnClearPreviousState(); // clear previous state
}
System.out.println("predict...");
double[] predicts = new double[n4];
for (int i = 0; i < n4; i++) {
predicts[i] = net.rnnTimeStep(dataSets2.get(i).getFeatureMatrix()).getDouble(n3 - 1) * (max1 - min1) + min1;
}
System.out.println("Show prediction");
List<Double> toShow1 = tprices1.subList(n1 - n4 * 2, n1 + n4);
List<Double> toShow2 = new ArrayList<>();
for (int i = 0; i < n4 * 2; i++) {
toShow2.add(toShow1.get(i));
}
for (int i = 0; i < n4; i++) {
toShow2.add(predicts[i]);
}
JFrame gp = new JFrame(symbol);
gp.add(new GraphPanel(toShow1, toShow2));
gp.setSize(800, 600);
gp.setDefaultCloseOperation(WindowConstants.DISPOSE_ON_CLOSE);
gp.setVisible(true);
}
