.gitignore

# =============================================================================
# Advanced RL Deep Dive — .gitignore
# =============================================================================

# ── Python ──────────────────────────────────────────────────────────────────
__pycache__/
*.py[cod]
*$py.class
*.so
*.pyd

# Distribution / Packaging
.Python
build/
dist/
eggs/
.eggs/
*.egg-info/
*.egg
MANIFEST
sdist/
wheels/
pip-wheel-metadata/
share/python-wheels/

# Virtual environments
.env
.venv
env/
venv/
ENV/
env.bak/
venv.bak/

# ── Jupyter Notebook ─────────────────────────────────────────────────────────
.ipynb_checkpoints/
*/.ipynb_checkpoints/
profile_default/
ipython_config.py

# ── NumPy / PyTorch / RL framework cache ─────────────────────────────────────
*.npy
*.npz
*.pkl
*.pickle
*.joblib
*.pt
*.pth
*.bin
*.safetensors
*.onnx
*.ckpt

# Framework caches (PyTorch, Stable-Baselines3, CleanRL)
.torch_cache/
.sb3_cache/
.cleanrl_cache/

# ── Gymnasium / MuJoCo cache ─────────────────────────────────────────────────
# MuJoCo binary / license files (구버전 호환)
mjkey.txt
mujoco200/
mujoco210/
mjpro150/
.mujoco/

# Gymnasium / Gym 환경 캐시
.gym_cache/
.gymnasium_cache/

# Atari ROM (legal 이슈로 절대 commit 금지)
*.bin
roms/
atari_roms/

# ── Replay buffer / 학습 데이터 ──────────────────────────────────────────────
# Replay buffer 직렬화 (Off-policy: SAC, TD3, DDPG, CQL, TD3+BC)
replay_buffer/
replays/
*.replay
*.buffer

# Trajectory / rollout 저장
trajectories/
rollouts/
demonstrations/

# Offline RL dataset (D4RL, RL Unplugged 등)
# Ch7-01 CQL · TD3+BC · BCQ · IQL 실험에서 사용
d4rl/
rl_unplugged/
offline_data/
*.h5
*.hdf5

# ── Experiment outputs ───────────────────────────────────────────────────────
# 생성된 시각화·학습 곡선·plot (각 챕터)
# Ch1: Performance Difference Lemma · Trust Region bound visualization · KL trajectory
# Ch2: TRPO 의 KL trajectory · CG iteration · line search backtrack 횟수
# Ch3: PPO clipping ratio 분포 · ε ablation · GAE λ 비교
# Ch4: SAC entropy plot · α 자동 튜닝 trajectory · soft Bellman 수렴
# Ch5: TD3 Q-value 과대추정 ablation · target smoothing 효과 · delayed update 비교
# Ch6: MuJoCo 5환경 × 5알고리즘 benchmark · sample efficiency curve
# Ch7: CQL OOD penalty · Dreamer world model rollout · MAML adaptation curve · DPO loss
*.png
*.jpg
*.jpeg
*.pdf
*.gif
*.mp4
*.webm
*.webp

# 훈련 결과 / 체크포인트
results/
outputs/
logs/
checkpoints/
runs/
wandb/
mlruns/
tensorboard/
tb_logs/
lightning_logs/
*.log
*.csv
*.tfevents.*

# 환경별 video recording (gymnasium video wrapper)
videos/
recordings/
*.mov
*.avi

# 재현 실험 산출물 (논문별)
reproductions/kakade2002_apprx/*.pkl
reproductions/schulman2015_trpo/*.pt
reproductions/schulman2017_ppo/*.pt
reproductions/haarnoja2018_sac/*.pt
reproductions/haarnoja2019_sac_auto/*.pt
reproductions/fujimoto2018_td3/*.pt
reproductions/lillicrap2016_ddpg/*.pt
reproductions/espeholt2018_impala/*.pt
reproductions/horgan2018_apex/*.pt
reproductions/kapturowski2019_r2d2/*.pt
reproductions/kumar2020_cql/*.pt
reproductions/fujimoto2021_td3bc/*.pt
reproductions/fujimoto2019_bcq/*.pt
reproductions/kostrikov2022_iql/*.pt
reproductions/hafner2019_dreamer/*.pt
reproductions/schrittwieser2020_muzero/*.pt
reproductions/finn2017_maml/*.pt
reproductions/duan2016_rl2/*.pt
reproductions/christiano2017_rlhf/*.pt
reproductions/ouyang2022_instructgpt/*.pt
reproductions/rafailov2023_dpo/*.pt

# Pretrained weights / reward models
pretrained/
weights/
reward_models/

# ── RLHF / LLM 관련 (Ch7-04) ─────────────────────────────────────────────────
# Reward model · preference dataset · SFT checkpoint
preference_data/
sft_checkpoints/
rm_checkpoints/

# HuggingFace cache (RLHF 실험에서 사용)
.cache/huggingface/
hub/
models--*/
datasets_cache/

# ── Hyperparameter sweeps ────────────────────────────────────────────────────
sweeps/
optuna_studies/
*.db
*.sqlite
*.sqlite3

# ── IDE / Editor ─────────────────────────────────────────────────────────────
# JetBrains (PyCharm, IntelliJ)
.idea/
*.iml
*.iws
*.ipr

# VSCode
.vscode/
*.code-workspace

# Vim / Emacs
*.swp
*.swo
*~
\#*\#
.\#*

# macOS
.DS_Store
.AppleDouble
.LSOverride
._*
.DocumentRevisions-V100
.fseventsd
.Spotlight-V100
.TemporaryItems
.Trashes
.VolumeIcon.icns
.com.apple.timemachine.donotpresent

# Windows
Thumbs.db
Thumbs.db:encryptable
ehthumbs.db
Desktop.ini
$RECYCLE.BIN/
*.lnk

# Linux
.directory

# ── Documentation build ──────────────────────────────────────────────────────
docs/_build/
docs/generated/
site/
.mkdocs_cache/

# ── Secrets / Credentials ────────────────────────────────────────────────────
.env.local
.env.*.local
secrets.py
credentials.json
*.pem
*.key
openai_api_key*
hf_token*
wandb_api_key*

# ── GPU / CUDA cache ─────────────────────────────────────────────────────────
.nv_cache/
.nvidia/
*.cu.o
.torch_extensions/

# ── Compiled extensions / build artifacts ────────────────────────────────────
*.ninja
*.ninja_log
*.ninja_deps

# ── Miscellaneous ────────────────────────────────────────────────────────────
.hypothesis/
.pytest_cache/
.mypy_cache/
.ruff_cache/
.coverage
htmlcov/
.tox/
*.mo
*.pot

# data/ 디렉토리 (학습용 데이터 보존)
data/
datasets/
raw/
processed/
!data/.gitkeep