55bab51a583edf62328c654f72321d3f0caf62a2,yarll/agents/sac.py,SAC,build_target_value_network,#SAC#Any#,152

Before Change

        x = self.states
        x = tf.nn.relu(tf.nn.xw_plus_b(x, target_net[0], target_net[1]))
        x = tf.nn.relu(tf.nn.xw_plus_b(x, target_net[2], target_net[3]))
        value = tf.nn.xw_plus_b(x, target_net[4], target_net[5], name="target_value")

        return value, target_update

After Change

        target_net = [ema.average(v) for v in value_vars]

        x = self.states
        for i in range(self.config["n_hidden_layers"]):
            x = tf.nn.relu(tf.nn.xw_plus_b(x, target_net[i * 2], target_net[i * 2 + 1]))
        value = tf.nn.xw_plus_b(x, target_net[i + 1], target_net[i + 2], name="target_value")

Instances