VictorMylle/RL-Sim-Framework
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

♻️ crazy refactor

Browse Source
This commit is contained in:
Victor Mylle
2026-03-11 22:52:01 +01:00
parent 35223b3560
commit 4115447022
34 changed files with 4255 additions and 102 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
configs/training/ppo.yaml
View File

@@ -12,5 +12,23 @@ entropy_loss_scale: 0.05
log_interval: 1000
checkpoint_interval: 50000
initial_log_std: 0.5
min_log_std: -2.0
max_log_std: 2.0
record_video_every: 10000
# ClearML remote execution (GPU worker)
remote: false
# ── HPO search ranges ────────────────────────────────────────────────
# Read by scripts/hpo.py — ignored by TrainerConfig during training.
hpo:
learning_rate: {min: 0.00005, max: 0.001}
clip_ratio: {min: 0.1, max: 0.3}
discount_factor: {min: 0.98, max: 0.999}
gae_lambda: {min: 0.9, max: 0.99}
entropy_loss_scale: {min: 0.0001, max: 0.1}
value_loss_scale: {min: 0.1, max: 1.0}
learning_epochs: {min: 2, max: 8, type: int}
mini_batches: {values: [2, 4, 8, 16]}