Release 1.6.0

Lightwood 1.6.0 changelog:

Many thanks to our community contributors for this release!
@MichaelLantz @mrandri19 @ongspxm @vaithak

Features:

• SHAP analysis block (#679, @mrandri19)
• Disable GlobalFeatureImportance when we have too many columns (#681, @ongspxm; #698)
• Added cleaner support for file path data types (image, audio, video) (#675)
• Add partial_fit() to sktime mixer (#689)
• Add ModeEnsemble (#692, @mrandri19)
• Add weighted MeanEnsembler (#680, @vaithak)

Bug fixes:

• Normalized column importance range (#690)
• Fix ensemble supports_proba in calibrate.py (#694, @mrandri19)
• Remove self-referential import (#696)
• Make a integration test for time_aim (#685, @MichaelLantz)
• Fix for various datasets (#700)

Other

• Improve logging for analysis blocks (#677; @MichaelLantz)
• Custom block example: LabelEncoder (#663)
• Implement ShapleyValues analysis (#679)
• Move array/TS normalizers to generic helpers (#702)

Release 1.5.0

Lightwood 1.5.0 changelog:

Many thanks to this month's community contributors!
@alteregoprofile, @LyndonFan, @MichaelLantz, @mrandri19, @ongspxm

Features:

• MFCC-based audio encoder (#625, #638; @mrandri19)
• Quantum mixer (#645, @ongspxm)
• Identity encoders (#623; @LyndonFan)
• Simpler default splitter (#624)
• MeanEnsemble (#658; @mrandri19)
• Improved interface to predict with all mixers (#627)
• API: predictor_from_json_ai (#633; @mrandri19)
• One-hot encoder mode to work without unknown categories (#639; @mrandri19)
• System for handling optional dependencies (#640)

Bug fixes:

• Img2Vec encoder bug fixes and tests (#619, #622; @mrandri19)
• Fix encoder prepare calls (#630)
• Black formatter fix (#650)
• Docs: doc_build triggers during pull_request (#653, #665; @MichaelLantz)
• ArrayEncoder fixes (#604, @alteregoprofile)

Other

• Rename fit_on_validation to fit_on_all (#626)
• Smaller test datasets (#631)
• Docs: add a time series forecasting tutorial (#635)
• Improved documentation coverage (#654, #660)
• Docs: doc_build automatically runs jupyter notebooks (#657)

Release 1.4.0

Lightwood 1.4.0 changelog:

Features:

• Streamlined dynamic .predict() argument passing (#563)
• Set default logging level with environment variable (@mrandri19, #603)
• Colored logs (@mrandri19, #608)

Bug fixes:

• JsonAI blocks are now Modules (#569)
• Ignore column drop error if column is not in the dataframe (#579)
• LightGBM dependency issue (#609)

Other

• Introduction to statistical analyzer tutorial (#577)
• Custom cleaner tutorial (#581)
• Custom mixer tutorial (#575)
• Custom analysis block tutorial (#576)
• Docstring for BaseEncoder (#587)
• Native Jupyter notebook support inside docs (#586)
• Automated docs deployment (#610)
• Updated CLA bot (#612)
• Improved README.md and CONTRIBUTING.md (#613)

Note: benchmarks will not run on the latest commit for this release, they were instead successfully ran for commit 79f27325a0877bb95709373007a97161fc9bb2eb .

Release 1.3.0

Lightwood 1.3.0 changelog:

Features:

• Modular Cleaner (#538 and #568)
• Modular Analysis (#539)
• Better Imports (#540)
• Improved Json AI default arguments (#543)
• Add seed to splitter (#553)
• Stratification and 3-way splitting (#542, #571)
• Use MASE metric for TS model selection (#499)

Bug fixes:

• Allow quantity as target (#546)
• Fix for LightGBM device check (#544)
• Select OneHotEncoder at Json AI build time and fix pd.None bugs (#549)
• Miscellaneous fixes (#570)

Other

• Improved CONTRIBUTING.md (#550)

Release v1.2.0

Features:

• Better defaults for Neural model in time series tasks (#461)
• Seed keyword passed (#482)
• Handle ' and " in dataset column names (#503)
• Helper function to split grouped time series (#501)
• Enhanced date-time + tag histograms (#502)
• Nonconformist speed optimizations (#497)
• Add dtype.tsarray (#530)

Bug fixes:

• Fix analysis memory usage (#485)
• Fix incorrect return value for order column in time series tasks (#488)
• Fix time series encoding issue (#495)
• Remove deprecated logic (#518)
• Make explainer work with categorical targets not present in the training data (#500)
• Fix sktime dependency (#524)
• Better detection, cleaning and encoding of arrays (#512)
• Use correct accuracy score for binary data (#532)
• allow_incomplete_history for time series predictors (#525)

Other

• Automated documentation (NOTE: still in beta; #519, #528)
• Rename model to mixer; folds to subsets (#534)