microsoft / TextWorld

​TextWorld is a sandbox learning environment for the training and evaluation of reinforcement learning (RL) agents on text-based games.
Other
1.23k stars 188 forks source link

1.2.state tracker wrapper #185

Closed MarcCote closed 5 years ago

MarcCote commented 5 years ago

This PR refactors the whole TextWorld Environments submodule. With this PR, separate wrappers have been added to deal with 1) extracting additional information and 2) doing state tracking. Those wrappers work independently from the type of game (.z8 or .ulx) as long as it is a game generated by TextWorld (i.e. there's a .json).

@tavianator I know it's a big PR. I'll try to break it down if I can, but if you can start looking at it so we can discuss it next week that would be awesome. Thanks.

MarcCote commented 5 years ago

Requires https://github.com/microsoft/jericho/pull/19