55bab51a583edf62328c654f72321d3f0caf62a2,yarll/agents/sac.py,SAC,build_actor_network,#SAC#,87

Before Change

        w_bound = 3e-3
        x = self.states
        with tf.variable_scope("actor"):
            x = tf.nn.relu(linear(x, self.config["n_hidden_units"], "L1", normalized_columns_initializer()))
            x = tf.nn.relu(linear(x, self.config["n_hidden_units"], "L2", normalized_columns_initializer()))

            mean = linear(x,
                          self.n_actions,
                          "mean",

After Change

        w_bound = 3e-3
        x = self.states
        with tf.variable_scope("actor"):
            for i in range(self.config["n_hidden_layers"]):
                x = tf.nn.relu(linear(x, self.config["n_hidden_units"], "L{}".format(i + 1), normalized_columns_initializer()))

            mean = linear(x,
                          self.n_actions,
                          "mean",
                          tf.random_uniform_initializer(-w_bound, w_bound),

Instances