Overview¶
open-mlpipe is a production-level automated ML pipeline that takes your raw dataset and automatically:
- Loads and profiles your data (EDA)
- Cleans duplicates, outliers, and missing values
- Engineers features (interactions, log transforms, missingness flags)
- Compares 14+ ML models head-to-head
- Tunes hyperparameters with Optuna
- Selects features via SHAP importance
- Explains predictions with SHAP
- Saves a production-ready inference pipeline
Architecture¶
Raw Data
↓
[1] Load & EDA
↓
[2] Clean
↓
[3] Feature Engineering
↓
[4] Split (train/test)
↓
[5] Preprocess (impute, scale, encode)
↓
[6] Compare 14+ Models
↓
[7] Tune Best Model (Optuna)
↓
[8] Select Features (SHAP)
↓
[9] Evaluate (test set)
↓
[10] Explain (SHAP)
↓
[11] Save (full pipeline)
↓
PRODUCTION MODEL READY
Key Features¶
| Feature | Description |
|---|---|
| Auto EDA | Column hygiene, quality, distributions, statistical tests, VIF |
| Feature Engineering | Missingness flags, interactions, log transforms, datetime |
| 14+ Models | Linear, tree, boosting, bagging, ensemble |
| Hyperparameter Tuning | Optuna with 18 model-specific search spaces |
| SHAP Explainability | Global + local feature importance |
| Production Inference | Full pipeline saved (feature engineering + model) |
| Overfitting Detection | Automatic train/test gap analysis |
| Production Metrics | R², RMSE, MAE, MAPE, F1, ROC-AUC, MCC |
Supported Tasks¶
- Regression — predict continuous values
- Classification — predict categories (binary/multiclass)