Can't get a simple XOR problem network to work, answer always array([0])

I am trying to implement a XOR-problem solving network and can’t seem to get it to work. Here’s my code:

model = Sequential()
model.add(Dense(2,2))
model.add(Activation('sigmoid'))
model.add(Dense(2,1))
model.add(Activation('softmax'))
X = numpy.array([[0,0],[0,1],[1,0],[1,1]])
y = numpy.array([[0],[1],[1],[0]])
model.compile(loss='categorical_crossentropy', optimizer='sgd')
model.fit(X, y, nb_epoch=5, batch_size=32)

No matter what input I try, the answer to predict_classes is always array([0]), with predict_proba result always being array([[ 1.]]).

I have tried other setups, with tanh activation, loss='mean_absolute_error', optimizer='rmsprop', nb_epoch=20, batch_size=16, but there was no difference.

[logloglog]

If I weren’t interested in learning how to use Keras, I wouldn’t have raised this issue. If you don’t have the time to help, then you shouldn’t spend it writing a passive-aggressive retort either.

[dcgrigsby]

Here’s an XOR net that works with the new API:

import numpy as np
from keras.models import Sequential
from keras.layers.core import Activation, Dense
from keras.optimizers import SGD

X = np.zeros((4, 2), dtype='uint8')
y = np.zeros(4, dtype='uint8')

X[0] = [0, 0]
y[0] = 0
X[1] = [0, 1]
y[1] = 1
X[2] = [1, 0]
y[2] = 1
X[3] = [1, 1]
y[3] = 0

model = Sequential()
model.add(Dense(2, input_dim=2))
model.add(Activation('sigmoid'))
model.add(Dense(1))
model.add(Activation('sigmoid'))

sgd = SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(loss='mean_squared_error', optimizer=sgd, class_mode="binary")

history = model.fit(X, y, nb_epoch=10000, batch_size=4, show_accuracy=True, verbose=0)

print model.predict(X)