[Python] CartPole-v1, 강화학습 행동을 막대 기울기에 의존해보기

https://github.com/ParkGyeongTae/dqn-atari-breakout/tree/main/2_cartpole-v1

GitHub - ParkGyeongTae/dqn-atari-breakout

 

### Step 1) 아나콘다 가상환경 만들기

conda create -n py37 python=3.7

### Step 2) 가상환경 접속

conda activate py37

### Step 3) 라이브러리 설치

pip install gym==0.23.1
pip install pygame==2.1.2
pip install tensorflow==2.8.0
pip install keras==2.8.0
pip install torch==1.11.0
pip install matplotlib==3.5.1

 

 

2_cartpole-v1-reward-degree.py

import gym
import random

env = gym.make('CartPole-v1')

max_time_step   = 100
episode_numbers = 10

for episode in range(episode_numbers):

  observation = env.reset()
  rewards = 0

  for time_step in range(max_time_step):

    env.render()

    if observation[2] > 0:
      action = 1
    elif observation[2] < 0:
      action = 0
    else:
      action = random.randrange(0, 2)

    observation, reward, done, info = env.step(action)

    rewards += reward

    if done:
      print("episode :", episode + 1, ", rewards :", rewards)
      break

env.close()

 

python 2_cartpole-v1-reward-degree.py

 

랜덤행동과 기울기에따른행동 비교