5. filter_shape:(number of filters, num input feature maps,filter height
6. image_shape:(batch size, num input feature maps,image height, image w
__init__(self, rng, input, filter_shape, image_shape, poolsi
ze=(2, 2)):
12.#assert condition,condition 为 True,则继续往下执行,condition 为 False,
13.#image_shape[1]和 filter_shape[1]都是 num input feature maps,它们必须是
image_shape[1] == filter_shape[1]
self.input = input
17.#每个隐层神经元(即像素)与上一层的连接数为
fan_in = numpy.prod(filter_shape[1:])
"num output feature maps * filter height * filter width" /pooling siz
e
fan_out = (filter_shape[0] * numpy.prod(filter_shape[2:]) /
numpy.prod(poolsize))
25.#以上求得 fan_in、fan_out ,将它们代入公式,以此来随机初始化 W,W就是线性卷积
W_bound = numpy.sqrt(6. / (fan_in + fan_out))
self.W = theano.shared(
评论0
最新资源