layers.py 文件源码

def __init__(self, rng, input, filter_shape, image_shape, activation,
                 padding, W=None, b=None, b_v=0., stride=(1, 1)):
        """Implement a convolution layer. No pooling."""
        assert image_shape[1] == filter_shape[1]
        self.input = input
        self.x = input
        print filter_shape, "***********"
        fan_in = numpy.prod(filter_shape[1:])
        fan_out = (filter_shape[0] * numpy.prod(filter_shape[2:]))
        # initialize weights with random weights
        W_bound = numpy.sqrt(6. / (fan_in + fan_out))
        if rng is None:
            rng = numpy.random.RandomState(23455)
        if W is None:
            W = theano.shared(
                numpy.asarray(
                    rng.uniform(low=-W_bound, high=W_bound, size=filter_shape),
                    dtype=theano.config.floatX
                ),
                name="w_conv",
                borrow=True
            )
        if b is None:
            b_v = (
                numpy.ones(
                    (filter_shape[0],)) * b_v).astype(theano.config.floatX)
            b = theano.shared(value=b_v, name="b_conv", borrow=True)

        self.W = W
        self.b = b
        conv_out = conv2d(
            input=self.x,
            filters=self.W,
            input_shape=image_shape,
            filter_shape=filter_shape,
            border_mode=padding,
            subsample=stride
        )
        linear = conv_out + self.b.dimshuffle('x', 0, 'x', 'x')
        if activation is not None:
            self.output = activation(linear)
        else:
            self.output = linear
        self.params = [self.W, self.b]