// @Test
public void test() throws Exception {
InputStream ftrn =
IOUtils.createFileInputStream("/Users/jdchoi/Documents/Data/mnist/mnist_trn.txt");
InputStream ftst =
IOUtils.createFileInputStream("/Users/jdchoi/Documents/Data/mnist/mnist_tst.txt");
boolean sparse = true;
List<Instance> trn = read(ftrn, sparse);
List<Instance> tst = read(ftst, sparse);
WeightVector w = new WeightVector();
OnlineOptimizer op;
// best: 89.87 at 46 epoch
// op = new Perceptron(w, 0.01f, 0);
// develop(trn, tst, op, 1, sparse);
// best: 90.64 at 41 epoch
// op = new AdaGrad(w, 0.01f, 0);
// develop(trn, tst, op, 1, sparse);
// best: 90.60 at 26 epoch
// op = new AdaGrad(w, 0.01f, 0, new RegularizedDualAveraging(w, 0.001f));
// develop(trn, tst, op, 1, sparse);
// best: 90.49 at 24 epoch
// op = new AdaGradMiniBatch(w, 0.01f, 0);
// develop(trn, tst, op, 5, sparse);
// best: 90.63 at 45 epoch
// op = new AdaGradMiniBatch(w, 0.01f, 0, new RegularizedDualAveraging(w, 0.001f));
// develop(trn, tst, op, 5, sparse);
// best: 89.12 at 24 epoch
// op = new AdaDeltaMiniBatch(w, 0.01f, 0.4f, 0);
// develop(trn, tst, op, 5, sparse);
// best: 89.36 at 7 epoch
// op = new AdaDeltaMiniBatch(w, 0.01f, 0.4f, 0, new RegularizedDualAveraging(w, 0.001f));
// develop(trn, tst, op, 5, sparse);
// best: 92.38 at 47 epoch
// op = new SoftmaxRegression(w, 0.00000001f, 0);
// develop(trn, tst, op, 1, sparse);
// best: 92.66 at 46 epoch
op = new AdaGradRegression(w, 0.0001f, 0);
develop(trn, tst, op, 1, sparse);
// ActivationFunction sigmoid = new SigmoidFunction();
// op = new FeedForwardNeuralNetworkSoftmax(new int[]{300}, new ActivationFunction[]{sigmoid},
// 0.0001f, 0, new RandomWeightGenerator(new XORShiftRandom(9), -0.2f, 0.2f));
// develop(trn, tst, op, 1, sparse);
}
void iterate(TSVReader<POSNode> reader, List<String> filenames, Consumer<POSNode[]> f)
throws Exception {
for (String filename : filenames) {
reader.open(IOUtils.createFileInputStream(filename));
POSNode[] nodes;
while ((nodes = reader.next()) != null) f.accept(nodes);
reader.close();
}
}