Daniel Addis

Several changes including working counts_map, modifiable to report to wandb less-frequently, some speed-related (debatable) changes, stats reporting put in different loop, config.yaml updated with best settings, etc.

Fixed Spaces env: dtype=np.int8 -> dtype=np.float32, which was causing model to fail to learn. Reset dtype updated for consistency.

added options=None to fix broken ocean envs + updated includes setup.py for raylib rendering and cffi

1

2 files changed only

Comments are in the commits. Note: doesn't train well locally.

1. [new bindings for connect4. tested locally.](https://github.com/PufferAI/PufferLib/commit/fa82260f44dde2439d748dc810d0964d9132c0d8) 2. [additional refactor. removed unnecessary passing of constants via python](https://github.com/PufferAI/PufferLib/commit/77937f5501670a0d91f34344f0a74ee9a55d481e)

That should be all of the changes and a few more. New model trained off current code.

env rendering as discussed