agent.py 文件源码

def _network(self):
        # with tf.variable_scope(scope):
        w_init = tf.random_normal_initializer(0., .1)
        # actor part
        # return mu & sigma to determine action_norm_dist
        scope_var = "actor"
        mu, sigma = net_frame.mlp_frame([200] , self.state , self.action_dim , scope_var, \
                                        activation_fn=tf.nn.relu6, w_init=w_init, activation_fn_v=tf.nn.tanh, \
                                        activation_fn_a=tf.nn.softplus, continu=True)
        # cirtic part
        # return value of the state
        scope_var = "critic"
        v = net_frame.mlp_frame([100], self.state, 1, scope_var, activation_fn=tf.nn.relu6)
        return mu, sigma, v


# ===============================================================
#                             DDPG Agent
# ===============================================================