MNIST_COVNET.py 文件源码

def __init__(self, input, filter_shape, image_shape, padding=(0, 0), 
                 stride=(1, 1), activation_fn=None, seed=3235):

        assert image_shape[1] == filter_shape[1]

        # rng = np.random.RandomState(seed)

        self.input = input
        self.filter_shape = filter_shape
        self.image_shape = image_shape
        self.activation_fn = activation_fn

        fan_in = np.prod(filter_shape[1:])
        fan_out = filter_shape[0]*np.prod(filter_shape[2:]) // 2
        W_bound = np.sqrt(6/(fan_in+fan_out))
        w = np.random.uniform(low=-W_bound, high=W_bound, size=filter_shape)
        b_vals = np.random.uniform(size=filter_shape[0])

        # Initiliaze weights with random variables
        self.W = theano.shared(name='weights',
                               value=w.astype(theano.config.floatX),
                               borrow=True)
        self.b = theano.shared(name='bias',
                               value=b_vals.astype(theano.config.floatX), 
                               borrow=True)

        conv_out = conv2d(input=input, filters=self.W, border_mode=padding,
                          subsample=stride, filter_shape=filter_shape, 
                          input_shape=image_shape)

        l_output = conv_out + self.b.dimshuffle(('x', 0, 'x', 'x'))
        self.output = (l_output if activation_fn is None 
                       else activation_fn(l_output))

        # Parameters of the model
        self.params = [self.W, self.b]