Default Artifact Paths for RL Examples #
TorchLean's runnable RL examples write small JSON artifacts to data/rl/ by default:
- training curves (
TrainLogJSON), - optional policy snapshots, and
- optional episode path snapshots.
Both the executable trainers (under NN/Examples/Models/RL) and the editor-side viewers (under
NN/Examples/RL) need to agree on these defaults. Keeping the strings here gives both sides one
shared path convention.
Users can always override these paths with the corresponding CLI flags (e.g. --log, --policy,
--path).
Default training-log path written by torchlean ppo_cartpole (override with --log).
Instances For
Default training-log path written by torchlean ppo_pong_ram (override with --log).
Instances For
Default training-log path written by torchlean ppo_gridworld (override with --log).
Instances For
Default policy snapshot path written by torchlean ppo_gridworld (override with --policy).
Instances For
Default episode-path snapshot path written by torchlean ppo_gridworld (override with --path).