PG-REINFORCE tensorflow 2.0

REINFORCE 算法实现

REINFORCE算法是策略梯度算法最原始的实现算法，这里采用tensorflow2.0进行实现

import tensorflow as tf
import gym
from matplotlib import pyplot as plt
import numpy as np


def PGReinforce_run(PGReinforce_agent=None, episode=1000):
    PGReinforce_agent = PGReinforce_agent.PGReinforce(n_actions=2, n_features=4)
    PGReinforce_agent.net_init()
    score = []
    env = gym.make('CartPole-v1')
    bias = 5
    for i_episode in range(episode):
        # 初始化，
        observation = env.reset()
        done = False
        t = 0
        while not done:
            env.render()
            action = PGReinforce_agent.choose_action(observation)
            PGReinforce_agent.traj_store(observation, action)
            observation_, reward, done, info = env.step(action)
            x, x_dot, theta, theta_dot = observation
            r2 = - abs(theta)*5
            # r1 = - abs(x)
            PGReinforce_agent.r_calculate(reward + r2)
            observation = observation_
            t += 1
        # PGReinforce_agent.loss_calculate()
        print("Episode finished after {} time steps".format(t + 1))
        score.append(t + 1)
        PGReinforce_agent.learn(5)
        if (i_episode + 1) % 100 == 0:
            plt.plot(score)  # 绘制波形
            # plt.draw()
            plt.savefig(f"RL_algorithm_package/img/pic_