Update README

Remove UI part of the code for the upload of the repo to Github.com (may add it back later on)

Fixed some pathing issues via os.path.join + separate eval_rolling_window() function

Unnecessary ValueError, if we are done with an Episode, we don't care that the next state will be wrong.

paths, docker rllib version and NaNs

Refactoring the train_appo/ppo.py files so they look a bit more similar, improvement of the evaluate graphs + recordkeeping of results

Update fixing a bug in the infer_volume_from_action function.

Changes to the plotting + evaluation functions.

Old code cleanup + refactor of the filenames of the limit_orders setup.

Implemented iterative training with the APPO + Improvements to visualizations (addition of speed of execution calculation)

Solved the bug with the training ending, the # of iterations is kept across sessions if a checkpoint is restored.

Added lines to restore the best checkpoint from the earliest session run in the iterative training

Implemented a carry out a rolling-window training-evaluation experiment over a given time period

TODO

Changes to accommodate %Y-%M-%D info into all timestamps and be able to train over different days all at once.