Lasso Regression with UV + Notebooks

Setup

1. Install dependencies and create the environment:

uv sync

2. Open Jupyter in the project environment:

uv run jupyter lab

Notebook order

1. notebooks/preprocess_train.ipynb
2. notebooks/predict_evaluate.ipynb

Diagnostics included

• CV diagnostics (Lasso alpha error path + learning curve) to check overfitting risk.
• Regression plots (actual vs predicted, residuals vs predicted).
• Feature importance plot from Lasso coefficients.
• Optional thresholded classification metrics derived from regression outputs.
• Calibration-style threshold sweep for threshold selection.

Configuration

• Main configuration lives in configs/project.yaml.
• Machine-generated outputs are written to artifacts/*.json and artifacts/*.csv.

Artifacts

• artifacts/model.joblib — trained pipeline
• artifacts/splits.json — train/test split indices
• artifacts/training_summary.json — selected alpha and CV summary
• artifacts/predictions.csv — test predictions
• artifacts/metrics.json — RMSE/MAE/R² on test set
• artifacts/classification_metrics.json — thresholded classification metrics

Latest experiment results

Regression metrics (test set)

Metric	Value
RMSE	52.9196
MAE	42.7930
R²	0.4714

Thresholded classification metrics (test set)

Threshold used: 140.5

Metric	Value
Accuracy	0.7753
Precision	0.7083
Recall	0.8500
F1	0.7727
ROC-AUC	0.8423

Embedded report visuals

Overfitting and CV checks

Regression performance

Feature importance

Classification analysis