Upgrade to use RLLib master (almost v1.3).#10

Open

Manuscrit wants to merge 15 commits into

upgrade_from_rllib_1_0_0_to_rllib_master

Commits on Apr 15, 2021

Upgrade to use RLLib master (almost v1.3).
Maxime Riché
committed
Use _compute_actions_helper instead of compute_actions in policies. Small fixes.
Maxime Riché
committed
Fixing some tests, _comute_action_helper, update_target
Maxime Riché
committed

Commits on Apr 16, 2021

Fix amTFT exploiter.
Maxime Riché
committed

Commits on Apr 20, 2021

Support LSTM in amTFT.
Maxime Riché
committed

Commits on Apr 27, 2021

Add meta game exp with LOLA-Exact.
Maxime Riché
committed
Add several options to stabilize the LOLA-PG training.
Maxime Riché
committed

Commits on Apr 28, 2021

Fix bug in cross_play evaluator.
Maxime Riché
committed
Add SSDMMCG vectorized.
Maxime Riché
committed

Commits on May 7, 2021

Fix amTFT: how the environment is used during the rollouts
Maxime Riché
committed
Change LOLA-Exact to make it work with discrete actions space larger than 2.
Maxime Riché
committed
Add stuff to the environments:
Maxime Riché
committed
Add meta game experiments with various meta solvers (like: alpha-rank, replicator dynamic)
Maxime Riché
committed

Commits on May 13, 2021