batch_network.py 文件源码

def build_shared_network(self):
        """
        This part contains trhe sharred params (conv layer) for both policy and value networks

        Returns the shared output
        """
        #from lasagne.layers import Conv2DLayer

        l_in = lasagne.layers.InputLayer(
            shape=(self.batch_size, self.history_length, self.img_height, self.img_width)
        )

        #l_in = lasagne.layers.ReshapeLayer(l_in, (1, self.history_length, self.img_height, self.img_width))

        #l_conv1 = dnn.Conv2DDNNLayer(
        l_conv1 = lasagne.layers.Conv2DLayer(
            incoming=l_in,
            num_filters=16,
            filter_size=(8, 8),
            stride=(4, 4),
            nonlinearity=lasagne.nonlinearities.rectify,
            W=lasagne.init.HeUniform(), # Defaults to Glorot
            b=lasagne.init.Constant(.1)
            #dimshuffle=True
        )

        #l1_out=l_conv1.get_output_shape_for((self.history_length, self.img_height , self.img_width))
        #print "L1:", l1_out

        l_conv2 = lasagne.layers.Conv2DLayer(
            incoming=l_conv1,
            num_filters=32,
            filter_size=(4, 4),
            stride=(2, 2),
            nonlinearity=lasagne.nonlinearities.rectify,
            W=lasagne.init.HeUniform(),
            b=lasagne.init.Constant(.1)
            #dimshuffle=True
        )

        #l2_out=l_conv2.get_output_shape_for(l1_out)
        #print "L2:", l2_out

        l_hidden1 = lasagne.layers.DenseLayer(
            incoming=l_conv2,
            num_units=256,
            nonlinearity=lasagne.nonlinearities.rectify,
            W=lasagne.init.HeUniform(),
            b=lasagne.init.Constant(.1)
        )

        return l_hidden1