Release 23.2.1.0

Changelog

Features

• M1 support #1091
• [ENH] Improved sktime-based mixer interface and customization #1099
• [ENH] Dynamic mixer weights for numerical ensembles #1100
• [ENH] Automated parallelism check for encoder training #1083
• [feat] add special dispatch for bestof ensemble #1107

Fixes

• [hotfix] remove repeated dep in reqs.txt #1097
• [fix] Improved nltk messaging and check for no available mixers #1101
• [fix] single mixer usage in tsse ensemble #1104
• [fix] Lazy analysis - only compute mixer predictions if required #1105
• [feat] lazy explain when there are no blocks in line #1108

Benchmarks

http://benchmarks.mindsdb.com:9107/compare/best/1062ee8021aacf42b6cf0381a519277b155e2620

Release 23.1.2.0

Changelog

Note: due to refactors, this version is not backwards compatible.

Features

N/A

Fixes

• Add requirements_dev.txt #1084

Other

• [Refactor] Move evaluation logic into separate package #1089
• Benchmarks

Release 22.12.2.0

Changelog

Features

• XGBoost mixer (#1066)
• Store global insights inside predictor object (#1074)
• Force infer_row in bulk ts predictions (#1075)

Fixes

Other

• Bump transformers to 4.21.0 (#1070)
• Bump sktime 0.14.0 (#1077)
• Lightgbm-based mixers are now optional (#1080)

Release 22.11.2.0

Changelog:

Features:

• [ENH] Custom output distributions in GluonTS mixer (#1042)
• [ENH] Restores py3.7 support (#1039)

Bug fixes:

• [Fix] Parameters to be optimized (#1035)

Others:

• [Maint] Migrate to type_infer (#1024)
• [Test] Improve time series test (#1041)

Release 22.10.4.0

Features

• Argument to override device setting #1004
• Implement a RandomForest mixer #1017
• [ENH] ICP: aggregated sum over horizon #995
• Add lightwood library version to predictor #1018
• Add GluonTS mixer #1019
• Improve random forest mixer #1031
• Simpler TS evaluation #1032

Fixes

• Added original amount of columns #1014
• Fix supported python version spec >=3.7,<3.10 #1015
• Fix wrong time-series alignment and add WA for StatsForcastAutoARIMA #1022
• Fix: workaround for StatsForecastAutoARIMA #1025
• Fix: gluonts improvements #1030

Other

• Add test for forecast offset alignment #1023
• Remove n-hits & prophet from default mixer list #1028

---

Thanks to @alexandre-dz-oscore @riadhlaabidi @jaredc07 @akhildevelops @adripo @bachng2017 for contributing to this release!

Release 22.9.1.0

Release 22.9.1.0

Features

• [ENH] PyOD analysis block (#983)
• [ENH] Infer offset in SkTime mixers (#989)

Fixes

• [Fix] minor idx error in fh setting (#991)
• [Fix] stacked ensemble agg_dim (#992)

---

Benchmarks

Release 22.8.1.0

Changelog

Features

• Conformal forecasting #862
• Improved JsonAI templating for analysis blocks #960
• Some feature importance changes #961

Fixes

• Default NHITS space params #956
• Add data limits to PFI block #963

Release 22.7.4.0

Changelog

Features

• [ENH] New defaults for sktime mixer, new AutoETS and AutoARIMA mixers (#946)
• [ENH] Enforce bounded_ts_accuracy as default (#950)

Fixes

• fix quick start guide #943 (thanks to @ameliatheamazin!)
• [FIX] kwarg setting for LightGBM GPU (#944)
• [FIX] Hotfix: use defaults in new groups (#952)

Other

• [FIX] New bias description (#947)
• [ENH] Message for missing values (#948)

Release 22.7.3.0

Features

• [ENH] Add tn_conf for categorical targets (#936)

Bug fixes

• Better sp detection (#935)
• Various small fixes (#940)

Others

• Change supported Python versions to 3.7 and 3.8 (#930)
• Use prophet 1.1 with wheels (#911, thanks @abitrolly !)
• order_by is now a single column #938

Release 22.7.2.0

Release 22.7.2.0

Benchmarks

Features

• Experimental N-HITS forecasting mixer (#886)
• Differencing blocks for time series tasks (#903)
• Linear tree for LightGBMArray mixer (#902)
• STL decomposition blocks for time series tasks (#907)

Fixes

• Restored statsforecast as default backend for ARIMA models (#904)

Other

N/A