2-layered MLP - xor problem

gistfile1.txt

from random import randint
import numpy as np
from pylearn2.datasets.dense_design_matrix import DenseDesignMatrix
from pylearn2.models import mlp
from pylearn2.termination_criteria import EpochCounter
from pylearn2.training_algorithms import sgd
import theano


def gen_xor_data():
    x_input = [[randint(0, 1), randint(0, 1)] for _ in range(1000)]
    y_target = []
    for a, b in x_input:
        if a + b == 1:
            y_target += [[1]]
        else:
            y_target += [[0]]
    return x_input, y_target


if __name__ == "__main__":
    # build training data
    x_input, y_target = gen_xor_data()
    np_x_input = np.array(x_input)
    np_y_target = np.array(y_target)
    dataset = DenseDesignMatrix(X=np_x_input, y=np_y_target)

    # build network layers
    hidden_layer = mlp.Sigmoid(layer_name="hidden", dim=2, irange=.1,
                               init_bias=1.)  #mlp == multi-layer perceptron / dim == array dimensions
    output_layer = mlp.Linear(1, "output", irange=.1)
    layers = [hidden_layer, output_layer]

    #init trainer - to train layers with stochastic gradient descent (SGD)
    trainer = sgd.SGD(learning_rate=.05, batch_size=10, termination_criterion=EpochCounter(400))

    #begin network
    neural_network = mlp.MLP(layers, nvis=2)  #setup a simple perceptron network with 2 input
    trainer.setup(neural_network, dataset)  #attach trainer to the neural network

    #begin training the network with dataset
    print "Begin training network.."
    while True:
        trainer.train(dataset=dataset)
        # neural_network.monitor.report_epoch()
        # neural_network.monitor()
        if not trainer.continue_learning(neural_network):
            print "Training ends."
            break

    #test network
    for _ in range(5):
        random_input = [randint(0, 1), randint(0, 1)]
        print "Testing network with %d,%d" % (random_input[0], random_input[1])
        np_random_input = np.array([random_input])
        print np_random_input
        print neural_network.fprop(theano.shared(np_random_input, name="inputs")).eval()
        print "------"