update
This commit is contained in:
JohnJim0816
@@ -5,34 +5,47 @@ Author: John
|
||||
Email: johnjim0816@gmail.com
|
||||
Date: 2020-11-22 23:21:53
|
||||
LastEditor: John
|
||||
LastEditTime: 2020-11-24 19:52:40
|
||||
LastEditTime: 2021-03-13 11:50:32
|
||||
Discription:
|
||||
Environment:
|
||||
'''
|
||||
import sys,os
|
||||
sys.path.append(os.getcwd()) # 添加当前终端路径
|
||||
from itertools import count
|
||||
import torch
|
||||
import os
|
||||
from torch.utils.tensorboard import SummaryWriter
|
||||
import datetime
|
||||
import gym
|
||||
from PolicyGradient.agent import PolicyGradient
|
||||
from common.plot import plot_rewards
|
||||
from common.utils import save_results
|
||||
|
||||
from env import env_init
|
||||
from params import get_args
|
||||
from agent import PolicyGradient
|
||||
from params import SEQUENCE, SAVED_MODEL_PATH, RESULT_PATH
|
||||
from utils import save_results,save_model
|
||||
from plot import plot
|
||||
def train(cfg):
|
||||
env,state_dim,n_actions = env_init()
|
||||
device = torch.device("cuda" if torch.cuda.is_available() else "cpu") # 检测gpu
|
||||
agent = PolicyGradient(state_dim,device = device,lr = cfg.policy_lr)
|
||||
SEQUENCE = datetime.datetime.now().strftime("%Y%m%d-%H%M%S") # 获取当前时间
|
||||
SAVED_MODEL_PATH = os.path.split(os.path.abspath(__file__))[0]+"/saved_model/"+SEQUENCE+'/' # 生成保存的模型路径
|
||||
if not os.path.exists(os.path.split(os.path.abspath(__file__))[0]+"/saved_model/"): # 检测是否存在文件夹
|
||||
os.mkdir(os.path.split(os.path.abspath(__file__))[0]+"/saved_model/")
|
||||
if not os.path.exists(SAVED_MODEL_PATH): # 检测是否存在文件夹
|
||||
os.mkdir(SAVED_MODEL_PATH)
|
||||
RESULT_PATH = os.path.split(os.path.abspath(__file__))[0]+"/results/"+SEQUENCE+'/' # 存储reward的路径
|
||||
if not os.path.exists(os.path.split(os.path.abspath(__file__))[0]+"/results/"): # 检测是否存在文件夹
|
||||
os.mkdir(os.path.split(os.path.abspath(__file__))[0]+"/results/")
|
||||
if not os.path.exists(RESULT_PATH): # 检测是否存在文件夹
|
||||
os.mkdir(RESULT_PATH)
|
||||
|
||||
class PGConfig:
|
||||
def __init__(self):
|
||||
self.train_eps = 300 # 训练的episode数目
|
||||
self.batch_size = 8
|
||||
self.lr = 0.01 # 学习率
|
||||
self.gamma = 0.99
|
||||
self.hidden_dim = 36 # 隐藏层维度
|
||||
|
||||
def train(cfg,env,agent):
|
||||
'''下面带pool都是存放的transition序列用于gradient'''
|
||||
state_pool = [] # 存放每batch_size个episode的state序列
|
||||
action_pool = []
|
||||
reward_pool = []
|
||||
''' 存储每个episode的reward用于绘图'''
|
||||
rewards = []
|
||||
moving_average_rewards = []
|
||||
log_dir=os.path.split(os.path.abspath(__file__))[0]+"/logs/train/" + SEQUENCE
|
||||
writer = SummaryWriter(log_dir) # 使用tensorboard的writer
|
||||
ma_rewards = []
|
||||
for i_episode in range(cfg.train_eps):
|
||||
state = env.reset()
|
||||
ep_reward = 0
|
||||
@@ -55,55 +68,22 @@ def train(cfg):
|
||||
action_pool = []
|
||||
reward_pool = []
|
||||
rewards.append(ep_reward)
|
||||
if i_episode == 0:
|
||||
moving_average_rewards.append(ep_reward)
|
||||
if ma_rewards:
|
||||
ma_rewards.append(
|
||||
0.9*ma_rewards[-1]+0.1*ep_reward)
|
||||
else:
|
||||
moving_average_rewards.append(
|
||||
0.9*moving_average_rewards[-1]+0.1*ep_reward)
|
||||
writer.add_scalars('rewards',{'raw':rewards[-1], 'moving_average': moving_average_rewards[-1]}, i_episode+1)
|
||||
writer.close()
|
||||
print('Complete training！')
|
||||
save_model(agent,model_path=SAVED_MODEL_PATH)
|
||||
'''存储reward等相关结果'''
|
||||
save_results(rewards,moving_average_rewards,tag='train',result_path=RESULT_PATH)
|
||||
plot(rewards)
|
||||
plot(moving_average_rewards,ylabel='moving_average_rewards_train')
|
||||
|
||||
def eval(cfg,saved_model_path = SAVED_MODEL_PATH):
|
||||
env,state_dim,n_actions = env_init()
|
||||
device = torch.device("cuda" if torch.cuda.is_available() else "cpu") # 检测gpu
|
||||
agent = PolicyGradient(state_dim,device = device,lr = cfg.policy_lr)
|
||||
agent.load_model(saved_model_path+'checkpoint.pth')
|
||||
rewards = []
|
||||
moving_average_rewards = []
|
||||
log_dir=os.path.split(os.path.abspath(__file__))[0]+"/logs/eval/" + SEQUENCE
|
||||
writer = SummaryWriter(log_dir) # 使用tensorboard的writer
|
||||
for i_episode in range(cfg.eval_eps):
|
||||
state = env.reset()
|
||||
ep_reward = 0
|
||||
for _ in count():
|
||||
action = agent.choose_action(state) # 根据当前环境state选择action
|
||||
next_state, reward, done, _ = env.step(action)
|
||||
ep_reward += reward
|
||||
state = next_state
|
||||
if done:
|
||||
print('Episode:', i_episode, ' Reward:', ep_reward)
|
||||
break
|
||||
rewards.append(ep_reward)
|
||||
if i_episode == 0:
|
||||
moving_average_rewards.append(ep_reward)
|
||||
else:
|
||||
moving_average_rewards.append(
|
||||
0.9*moving_average_rewards[-1]+0.1*ep_reward)
|
||||
writer.add_scalars('rewards',{'raw':rewards[-1], 'moving_average': moving_average_rewards[-1]}, i_episode+1)
|
||||
writer.close()
|
||||
print('Complete evaling！')
|
||||
|
||||
ma_rewards.append(ep_reward)
|
||||
print('complete training！')
|
||||
return rewards, ma_rewards
|
||||
|
||||
if __name__ == "__main__":
|
||||
cfg = get_args()
|
||||
if cfg.train:
|
||||
train(cfg)
|
||||
eval(cfg)
|
||||
else:
|
||||
model_path = os.path.split(os.path.abspath(__file__))[0]+"/saved_model/"
|
||||
eval(cfg,saved_model_path=model_path)
|
||||
cfg = PGConfig()
|
||||
env = gym.make('CartPole-v0') # 可google为什么unwrapped gym，此处一般不需要
|
||||
env.seed(1) # 设置env随机种子
|
||||
n_states = env.observation_space.shape[0]
|
||||
n_actions = env.action_space.n
|
||||
agent = PolicyGradient(n_states,cfg)
|
||||
rewards, ma_rewards = train(cfg,env,agent)
|
||||
agent.save_model(SAVED_MODEL_PATH)
|
||||
save_results(rewards,ma_rewards,tag='train',path=RESULT_PATH)
|
||||
plot_rewards(rewards,ma_rewards,tag="train",algo = "Policy Gradient",path=RESULT_PATH)
|
||||
|
||||
Reference in New Issue
Block a user