The practitioner's forecasting library
-
Updated
Oct 12, 2026 - Python
The practitioner's forecasting library
Mase JS is a new way to write HTML entirely in your JavaScript.
Time Series Forecasting Models: ETS, ARIMA
BERT, torch-lightning, rnnlm, attention, mase etc.. for study
Download Yahoo stock price data and forecast their future values with a vector autoregressive (VAR) model. Compare to Naive model using MASE. Version control with DVC
Seven-day-ahead daily demand forecasting with walk-forward validation. MASE, calibrated prediction intervals, and the evaluation defects I found and pinned with tests.
Time-series forecasting benchmark: sktime classical + ML models backtested with expanding-window temporal CV (MASE/sMAPE), leakage-safe feature engineering, and a quality gate. Fully offline, reproducible.
Pharma demand forecasting with an honest rolling-origin backtest: SARIMA and ETS vs a seasonal-naive baseline across 20 synthetic SKUs. Per-SKU champion selection on validation folds, untouched holdout reporting (27.5% MASE gain; baseline wins 4 SKUs), and service-level cost framing. Deterministic, seeded, fully reproducible.
Week-ahead sales forecasting for a UK sports bar — synthetic data, MASE-scored rolling-origin CV, honest model comparison (SARIMA / Prophet / LSTM).
Seals a forecast before the outcome exists, grades it when reality arrives, and refuses to forecast when the history cannot support the evaluation protocol. Real Eurostat data, rolling-origin backtest, MASE, no hindsight.
Italian Environmental Remediation & Brownfield Feed — Programmatic REST API for National Priority contaminated sites (SIN), brownfield repowering, and solar eligibility.
To associate your repository with the mase topic, visit your repo's landing page and select "manage topics."