policy_gradient_actor_critic.py 文件源码

def policy_gradient():
    with tf.variable_scope("policy"):
        params = tf.get_variable("policy_parameters", [4, 2])
        state = tf.placeholder("float", [None, 4])
        actions = tf.placeholder("float", [None, 2])
        advantages = tf.placeholder("float", [None, 1])
        reward_input = tf.placeholder("float")
        episode_reward = tf.get_variable("episode_reward", initializer=tf.constant(0.))
        episode_reward = reward_input
        linear = tf.matmul(state, params)
        probabilities = tf.nn.softmax(linear)
        good_probabilities = tf.reduce_sum(tf.mul(probabilities, actions), reduction_indices=[1])
        eligibility = tf.log(good_probabilities) * advantages
        loss = -tf.reduce_sum(eligibility)
        optimizer = tf.train.AdamOptimizer(learning_rate).minimize(loss)

        tf.scalar_summary("loss", loss)
        tf.scalar_summary("episode_reward", episode_reward)
        return probabilities, state, actions, advantages, optimizer, reward_input, episode_reward