Nixtla/statsforecast v1.5.0

on GitHub

What's Changed

Features

New models

• [FEAT] ARIMA model (no auto version) in #383
• [FEAT] AutoRegressive model in #387
• [FEAT] GARCH and ARCH models in #403

New functionality

Forward methods

Now you can pre-train a model and use new data to make forecasts through the forward method. Supported models:

• [FEAT] Add forward method to Theta models in #362
• [FEAT] Add forward method to ETS models in #363
• [FEAT] Add forward method to AutoCES class in #364
• [FEAT] Add forward method to MSTL model in #369
• [FEAT] Add forward method to AutoARIMA (ARIMA and AutoRegressive) in #368

Misc

• [FEAT] Add alias argument to models (fit the same instance of models with different names) in #357
• [FEAT] Add cross-validation without refit (using the forward method) in #370
• [FEAT] Allow seasonality greater than 24 for ETS in #384
• [FEAT] Allow passing fixed coefficients for Arima in #386
• [FEAT] AutoCES prediction intervals in #394 (now StatsForecast is fully probabilistic)
• [FEAT] Add cla workflow in #351
• [FEAT] Add pypi downloads badge in #352
• [FEAT] Ignore jupyter notebooks as part of languages in #356
• [FEAT] Add nbdev merge to git attributes in #365
• [FEAT] Add citation in #366
• [FEAT] Update table of models in #396

Experiments

• [FEAT] Add M5 and M4-Daily experiments (Amazon Forecast) in #332
• [FEAT] Test recover M3 performance in #385
• [FEAT] BigQuery comparison in #421
• [FEAT] Experiments for ETS prediction intervals for multiple confidence levels in #377
• [FEAT] Add M3 experiment in #348
• [FEAT] Add a test ensuring the m3 performance is recovered in less than two minutes in #388

Tutorials

• [FEAT] Improved intermittent data nb in #359
• [FEAT] Add statistical and neural methods tutorial in #399
• [FEAT] Improve anomaly detection nb in #338
• [FEAT] GARCH and ARCH models tutorial in #418
• [FEAT] Improved notebook on prediction intervals in #358
• [FEAT] Improved notebook on exogenous regressors in #392
• [FEAT] Improve documentation in #376

Fixes

• [FIX] Exponential Smoothing description in #346
• [FIX] Changed dataset and model to make example easier to follow in #345
• [FIX] Readme M3 typo in #350
• [FIX] Delete CLA.yml in #355
• [FIX] Broken link in #360
• [FIX] Clean aws nbs in #361
• [FIX] Add correct link to hierarchicalforecast by #372
• [FIX] Recover table-based documentation (core nb, compatible with docstrings) in #374
• [FIX] update sklearn -> scikit-learn in #375
• [FIX] Ray CI in #381
• [FIX] Links and typos in documentation in #390
• [FIX] Correct evaluation using Winkler score by @MMenchero in #395
• [FIX] Recover plots prediction intervals tutorial in #398
• [FIX] Use https links instead of s3 uris (stat-neural tutorial) in #400
• [FIX] New nbdev clean behaviour in #412
• [FIX] Model imports in #408

New dependencies

• [FEAT] plotly-resampler as plotting engine in #354
• [FEAT] Move Fugue to core dependency in #419

New Contributors

• @jvdd made their first contribution in #354
• @Roymprog made their first contribution in #390
• @nelsoncardenas made their first contribution in #408

Full Changelog: v1.4.0...v1.5.0