search

Kautenja / playing-mario-with-deep-reinforcement-learning

An implementation of (Double/Dueling) Deep-Q Learning to play Super Mario Bros.
MIT License
68 stars 12 forks source link

Play Time Limit #11

Closed Kautenja closed 6 years ago

Kautenja commented 6 years ago

gym has a time limit built in https://github.com/openai/gym/blob/master/gym/wrappers/time_limit.py

Kautenja commented 6 years ago

5m = 18000 frames

Kautenja commented 6 years ago

180,000 frames is an epoch (50m of play time)

Kautenja commented 6 years ago

open ai gym environments have a 100,000 step (frame) by default

EDIT: 100,000 not 10,000. this is a good upper bound for a game (27.7 minutes). A TimeLimit on evaluation games that tightens the bound to 18,000 is still a good idea though. i.e. in play.py and play.ipynb. leaving issue open until this is implemented.

EDIT: filenames are play, not train

Kautenja commented 6 years ago

new -v10 environment spec for Atari games implements a 18,000 frame limit for games. It's implemented in play.py and play.ipynb. closing issue