In [1]:

    

import theano
from theano import tensor as T
import numpy as np
from load import mnist
from scipy.misc import imsave


    

In [2]:

    

def floatX(X):
    return np.asarray(X, dtype=theano.config.floatX)

def init_weights(shape):
    return theano.shared(floatX(np.random.randn(*shape) * 0.01))

def sgd(cost, params, lr=0.05):
    grads = T.grad(cost=cost, wrt=params)
    updates = []
    for p, g in zip(params, grads):
        updates.append([p, p - g * lr])
    return updates

def model(X, w_h, w_o):
    h = T.nnet.sigmoid(T.dot(X, w_h))
    pyx = T.nnet.softmax(T.dot(h, w_o))
    return pyx


    

In [3]:

    

trX, teX, trY, teY = mnist(onehot=True)

X = T.fmatrix()
Y = T.fmatrix()

# hidden layer
w_h = init_weights((784, 625))
# output layer
w_o = init_weights((625, 10))

py_x = model(X, w_h, w_o)
y_x = T.argmax(py_x, axis=1)


    

In [4]:

    

cost = T.mean(T.nnet.categorical_crossentropy(py_x, Y))
params = [w_h, w_o]
updates = sgd(cost, params)


    

In [5]:

    

train = theano.function(inputs=[X, Y], outputs=cost, updates=updates, allow_input_downcast=True)
predict = theano.function(inputs=[X], outputs=y_x, allow_input_downcast=True)


    

In [6]:

    

for i in range(10):
    # batch : 128
    for start, end in zip(range(0, len(trX), 128), range(128, len(trX), 128)):
        cost = train(trX[start:end], trY[start:end])
    print i, np.mean(np.argmax(teY, axis=1) == predict(teX))


    

    




0 0.7028
1 0.8295
2 0.8665
3 0.8822
4 0.8893
5 0.8947
6 0.8983
7 0.9019
8 0.9048
9 0.9071