[深度学习框架] Theano上回归神经网络实现

# coding: utf-8

import numpy as np
import theano.tensor as T
import matplotlib.pyplot as plt
import theano
from theano import function

class Layer(object):
    def __init__(self, inputs, input_size, output_size, activiation_function=None):
        self.W = theano.shared(np.random.normal(0, 1, (input_size, output_size)))
        self.b = theano.shared(np.zeros((output_size,)) + 0.1)
        self.y = theano.dot(inputs, self.W) + self.b
        self.activiation_function = activiation_function
        if activiation_function is None:
            self.outputs = self.y
        else:
            self.outputs = self.activiation_function(self.y)

# make up data
x_data = np.linspace(-1, 1, 300)[:, np.newaxis]
noise = np.random.normal(0, 0.05, x_data.shape)
y_data = np.square(x_data) - 0.5 + noise

# show the data
#plt.scatter(x_data, y_data)
#plt.show()

# determine the inputs dtype
x = T.dmatrix('x')
y = T.dmatrix('y')

# add layers
l1 = Layer(x, 1, 10, T.nnet.softplus)
l2 = Layer(l1.outputs, 10, 1, None)

# compute the cost
cost = T.mean(T.square(l2.outputs - y))

# comput the gradients
gW1, gb1, gW2, gb2 = T.grad(cost, [l1.W, l1.b, l2.W, l2.b])

# apply gradient descent
learning_rate = 0.05
train = function(
    inputs=[x, y], 
    outputs=cost, 
    updates=[(l1.W, l1.W-learning_rate*gW1), 
             (l1.b, l1.b-learning_rate*gb1),
             (l2.W, l2.W-learning_rate*gW2),
             (l2.b, l2.b-learning_rate*gb2)]
)

# predict
predict = function(inputs=[x], outputs=l2.outputs)

# plot the data
#fig = plt.figure()
#ax = fig.add_subplot(1, 1, 1)
#ax.scatter(x_data, y_data)
#plt.ion()
#plt.show()

for i in range(1000):
    # training
    err = train(x_data, y_data)
    if i%50 == 0:
        print i, err
        #try:
        #    ax.lines.remove(lines[0])
        #except Exception:
        #    pass
        #predict_y = predict(x_data)
        #lines = ax.plot(x_data, predict_y, 'r-', lw=5)
        #plt.pause(1)


fig = plt.figure()
ax = fig.add_subplot(1, 1, 1)
ax.scatter(x_data, y_data)
predict_y = predict(x_data)
lines = ax.plot(x_data, predict_y, 'r-', lw=5)