Minimum Chainer implementation of 3-Layer Perceptron for solving XOR

chainer_xor

#!/usr/bin/env python
import numpy as np
from chainer import cuda, Function, FunctionSet, gradient_check, Variable, optimizers
import chainer.functions as F

x_train = np.array([
    [[0., 0.]],
    [[0., 1.]],
    [[1., 0.]],
    [[1., 1.]]
], dtype=np.float32)

y_train = np.array([[0], [1], [1], [0]], dtype=np.int32)

model = FunctionSet(
    l1=F.Linear(2, 2),
    l2=F.Linear(2, 1)
)

def forward(x_data, y_data):
    x = Variable(x_data)
    t = Variable(y_data)
    h = F.sigmoid(model.l1(x))
    y = F.sigmoid(model.l2(h))
    return F.mean_squared_error(y, t)

optimizer = optimizers.MomentumSGD()
optimizer.setup(model.collect_parameters())

mean_loss = 0
last_epoch = 0

for epoch in xrange(10000):
    sum_loss = 0
    for i in xrange(len(x_train)):
        optimizer.zero_grads()
        loss = forward(x_train[i], y_train[i])
        sum_loss += loss.data
        loss.backward()
        optimizer.update()

    mean_loss = sum_loss / len(x_train)
    last_epoch = epoch + 1

    if last_epoch % 1000 == 0:
        print "Epoch %d\tLoss %f" % (last_epoch, mean_loss)

for i in xrange(len(x_train)):
    x = Variable(x_train[i])
    h = F.sigmoid(model.l1(x))
    y = F.sigmoid(model.l2(h))
    loss = forward(x_train[i], y_train[i])
    print loss.data, y_train[i], y.data

print last_epoch, mean_loss