Class MLP

examples/reinforcement_learning/tutorial_C51.py:93–105

# ##############################  Network  ####################################
class MLP(tl.models.Model):

    def __init__(self, name):
        super(MLP, self).__init__(name=name)
        self.h1 = tl.layers.Dense(64, tf.nn.tanh, in_channels=in_dim[0], W_init=tf.initializers.GlorotUniform())
        self.qvalue = tl.layers.Dense(
            out_dim * atom_num, in_channels=64, name='q', W_init=tf.initializers.GlorotUniform()
        )
        self.reshape = tl.layers.Reshape((-1, out_dim, atom_num))

    def forward(self, ni):
        qvalues = self.qvalue(self.h1(ni))
        return tf.nn.log_softmax(self.reshape(qvalues), 2)


class CNN(tl.models.Model):