search

danijar / dreamerv2

Mastering Atari with Discrete World Models
https://danijar.com/dreamerv2
MIT License
898 stars 195 forks source link

improve installation and usage #10

Closed AnthonyPoschen closed 3 years ago

AnthonyPoschen commented 3 years ago

A friend of mine wanting to learn ML asked me to help install this project. This pr aims to reduce complexity in usage and ensure it will always work when trying to install from any machine and run. The dockerfile also acts as an instructional for anyone who instead wishes to install locally.

All pip requirements are captured now in requirements.txt you can generate this from a pip env with "pip freeze" to maintain it in the future i was doing

pip freeze > requirements.txt

the hidden steps are now easily viewable to people from within the dockerfile to install all weird things needed to run the example such as the roms.

hopefully this makes this project much easier for less programmer minded people to pick this up and play with more easily.

I did comment out the assert around GPU usage so it can be used from a CPU also increasing availability for people to use.

AnthonyPoschen commented 3 years ago

to test the docker build as this is not yet merged you would need to instead run

docker build -t DreamerV2 git://github.com/zanven42/DreamerV2#main

danijar commented 3 years ago

Thanks, that's a great idea! I added a complete Dockerfile for both Atari and MuJoCo to the repository. Closing this PR.