deterministic_discrete_mlp_q_function.py 文件源码

def init_policy(self):
        output_vec = L.get_output(self._output_vec_layer, deterministic=True)
        action = tf.to_int64(tf.argmax(output_vec, 1))
        action_vec = tf.one_hot(action, self._n)
        max_qval = tf.reduce_max(output_vec, 1)

        self._f_actions = tensor_utils.compile_function([self._obs_layer.input_var], action)
        self._f_actions_vec = tensor_utils.compile_function([self._obs_layer.input_var], action_vec)
        self._f_max_qvals = tensor_utils.compile_function([self._obs_layer.input_var], max_qval)