pendulum_ddpg.py

import gym
from agent.continuous.seperate.ddpg import DDPG
from example_model.policy.mlp.continuous import MLPDDPGContinuousActor
from example_model.policy.mlp.continuous import MLPDDPGContinuousCritic
from agent import utils
import tensorflow as tf
import numpy as np
from tensorboardX import SummaryWriter

writer = SummaryWriter()

state_size = 3
output_size = 1
env = gym.make('Pendulum-v0')
sess = tf.Session()
target_actor = MLPDDPGContinuousActor('target_actor',state_size,output_size)
target_critic = MLPDDPGContinuousCritic('target_critic',state_size,output_size)
actor = MLPDDPGContinuousActor('actor',state_size,output_size)
critic = MLPDDPGContinuousCritic('critic',state_size,output_size)
agent = DDPG(sess,state_size,output_size,1,1,target_actor,target_critic,actor,critic)
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver()
#saver.restore(sess,'pendulum_ddpg/model')

ep_len = 200
ep = 0
clip = 2.0
epsilon = 1.0
while True:
    ep += 1
    state = env.reset()
    done = False
    total_state, total_reward, total_done, total_next_state, total_action = [], [], [], [], []
    score = 0
    for t in range(ep_len):
        #env.render()
        action = agent.get_action([state], epsilon)

        action = action[0]
        next_state, reward, done, _ = env.step(clip * action)

        score += reward

        agent.get_sample(state,action,reward,next_state,done)

        state = next_state
        if len(agent.memory) >= 1000:
            agent.train_model()

    agent.noise_generator.reset()
    if len(agent.memory) >= 1000:
        epsilon *= 0.995
    if ep < 300:
        print(ep, score, epsilon)
        writer.add_scalar('data/reward', score, ep)
        saver.save(sess, 'pendulum_ddpg/model')