mario-rl-tutorial's Introduction

mario-rl-tutorial

Gym Super Mario is an environment bundle for OpenAI Gym

Installation

pip3 install git+https://github.com/chris-chris/gym-super-mario
pip3 install git+https://github.com/openai/baselines

To load and run the environments, run

import gym
import ppaquette_gym_super_mario
env = gym.make('ppaquette/SuperMarioBros-1-1-v0')

To train the model, run

python3 train.py

Paramters

You can also customize the training with parameters.

python3 train.py --algorithm=deepq --timesteps=2000000 --log=stdout --env=ppaquette/SuperMarioBros-1-1-v0

	Description	Default	Parameter Type
env	Gym Environment	ppaquette/SuperMarioBros-1-1-v0	string
log	logging type : tensorboard, stdout	stdout	string
algorithm	Currently, support 2 algorithms : deepq, acktr	deepq	string
timesteps	Total training steps	2000000	int
exploration_fraction	exploration fraction	0.5	float
prioritized	Whether using prioritized replay for DQN	False	boolean
dueling	Whether using dueling network for DQN	False	boolean
num_cpu	number of agents for A3C(acktr)	4	int

mario-rl-tutorial's People

Contributors

Stargazers

Watchers

mario-rl-tutorial's Issues

ImportError: cannot import name 'conv'

Hello, I am somewhat of a beginner trying this out :)

When I initially run the python train.py

I get this error:
File "/home/benbartling/Desktop/mario-rl-tutorial/acktr/policies.py", line 3, in <module> from baselines.acktr.utils import conv, fc, dense, conv_to_fc, sample, kl_div ImportError: cannot import name 'conv'

Any tips greatly appreciated... Also is it possible to render the game play or even the training?? How long does the training take I am running on CPU... Is there a preset amount of episodes in training?

Thank you for your time in responding as well as creating this 👍

Training speed after 1st episode.

It seems that the game speed is fast(TIME goes fast) in the 1st episode,
but after 1st episode, game speed is slow(TIME goes slow),
So from the second episode, it cost lots of time to train.
Is that something wrong in wapper or in the game itself ?