artifacts/
├── v1/
│   ├── models/
│   │   ├── classical/   # Ridge, Lasso, ElasticNet, KNN ×3, SVR, XGBoost, LightGBM, RF
│   │   └── deep/        # Vanilla LSTM, Bi-LSTM, GRU, Attention-LSTM, TFT,
│   │                    # BatteryGPT, iTransformer, Physics-iTransformer,
│   │                    # DG-iTransformer, VAE-LSTM
│   └── scalers/         # MinMax, Standard, Linear, Sequence scalers
└── v2/
    ├── models/
    │   ├── classical/   # Same family + Extra Trees, Gradient Boosting, best_rul_model
    │   └── deep/        # Same deep models re-trained on v2 feature set
    ├── scalers/         # Per-model feature scalers
    └── results/         # Validation JSONs

Model Performance Summary (v3)

Rank	Model	R²	MAE	Family
1	XGBoost	0.9866	1.58	Classical
2	GradientBoosting	0.9860	1.38	Classical
3	LightGBM	0.9826	1.98	Classical
4	RandomForest	0.9814	1.83	Classical
5	ExtraTrees	0.9701	3.20	Classical
6	TFT	0.8751	3.88	Transformer
7	Weighted Avg Ensemble	0.8991	3.51	Ensemble

Usage

These artifacts are automatically downloaded by the Space on startup via scripts/download_models.py. You can also use them directly:

python

from huggingface_hub import snapshot_download

local = snapshot_download(
    repo_id="NeerajCodz/aiBatteryLifeCycle",
    repo_type="model",
    local_dir="artifacts",
    token="<your-token>",   # only needed if private
)

Framework

Classical models: scikit-learn / XGBoost / LightGBM .joblib
Deep models (PyTorch): .pt state-dicts (CPU weights)
Deep models (Keras): .keras SavedModel format
Scalers: scikit-learn .joblib

Citation

bibtex

@misc{aiBatteryLifeCycle2025,
  author  = {Neeraj},
  title   = {AI Battery Lifecycle — SOH/RUL Prediction},
  year    = {2025},
  url     = {https://huggingface.co/spaces/NeerajCodz/aiBatteryLifeCycle}
}