jeakwon/gym-numpad

gym-numpad

Install

pip install -U git+https://github.com/jeakwon/gym-numpad

gym-numpad test

import gym
import gym_numpad

env=gym.make('numpad2x2_test-v0')
for episode in range(2):
    env.reset()
    t, score = 0, 0
    while True:
        action = env.action_space.sample()
        obs, reward, done, info =env.step(action)
        env.render(mode='human')
        t+=1
        score+=reward
        print(t, score, obs, reward, done, info, end='\r')
        if done:
            break

gym-numpad + sb3 RecurrentPPO

import gym
import gym_numpad
from sb3_contrib import RecurrentPPO

model = RecurrentPPO("MlpLstmPolicy", 'numpad2x2_test-v0', verbose=1)
env = model.get_env()

for episode in range(5):
    t, score, done, lstm_states = 0, 0, False, None
    obs = env.reset()
    while True:
        action, lstm_states = model.predict(obs, state=lstm_states)
        obs, reward, done, info = env.step(action)
        
        t+=1
        score+=reward
        print(t, score, obs, reward, done, info, end='\r')
        # env.render(mode='human')
        if done:
            break

model.learn(10000)

for episode in range(5):
    t, score, done, lstm_states = 0, 0, False, None
    obs = env.reset()
    while True:
        action, lstm_states = model.predict(obs, state=lstm_states)
        obs, reward, done, info = env.step(action)
        
        t+=1
        score+=reward
        # env.render(mode='human')
        print(t, score, obs, reward, done, info, end='\r')
        if done:
            print(t, score, obs, reward, done, info)
            break

gym-numpad + sb3 RecurrentPPO + n_envs

import gym_numpad
import numpy as np
from sb3_contrib import RecurrentPPO
from stable_baselines3.common.env_util import make_vec_env


n_envs = 4
env = make_vec_env("numpad2x2_test-v0", n_envs=n_envs)
model = RecurrentPPO("MlpLstmPolicy", env, verbose=1)

env = model.get_env()
obs = env.reset()

lstm_states = None
episode_starts = np.ones((n_envs,), dtype=bool)
while True:
    action, lstm_states = model.predict(obs, state=lstm_states, episode_start=episode_starts, deterministic=False)
    obs, rewards, dones, info = env.step(action)
    episode_starts = dones
    env.render()

gym-numpad + sb3 RecurrentPPO + n_envs + training

import gym
import gym_numpad

import numpy as np
from sb3_contrib import RecurrentPPO
from stable_baselines3.common.env_util import make_vec_env

# Parallel environments
n_envs=4
env = make_vec_env("numpad2x2_test-v0", n_envs=n_envs)

model = RecurrentPPO("MlpLstmPolicy", env, verbose=1)
model.learn(total_timesteps=25000)
model.save("rppo_numpad2x2_test")

del model # remove to demonstrate saving and loading

model = RecurrentPPO.load("rppo_numpad2x2_test")

obs = env.reset()

lstm_states = None
episode_starts = np.ones((n_envs,), dtype=bool)
while True:
    action, lstm_states = model.predict(obs, state=lstm_states, episode_start=episode_starts, deterministic=False)
    obs, rewards, dones, info = env.step(action)
    episode_starts = dones
    env.render()