mase

The practitioner's forecasting library

Mase JS is a new way to write HTML entirely in your JavaScript.

Time Series Forecasting Models: ETS, ARIMA

A list of websites and projects built using masejs.

BERT, torch-lightning, rnnlm, attention, mase etc.. for study

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.

Honest time series evaluation: rolling-origin backtesting, MASE against a seasonal naive baseline, and prediction intervals checked for actual coverage (they mostly fail).

Forecasting models to predict annual live births in England & Wales.

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

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).