Skip to content

Latest commit

 

History

History
232 lines (146 loc) · 3.67 KB

CHANGELOG.md

File metadata and controls

232 lines (146 loc) · 3.67 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog.

[2023.3.11]

Changed

  • opt LVL

[2023.3.9]

Changed

  • Update dependencies

Fix

  • ValueError: X and y both have indexes, but they do not match.

[1.3.10]

Fix

  • TypeError in data_prepare Outliers filter

[1.3.9]

ADD

  • Up score AutoML (Blend best top5 models in AutoML)

[1.3.8]

ADD

  • optimization DataPreproc parametrs in BestSingleModel
  • rebuild AutoML pepline (light version)

Fix

  • target encodet only cat features

[1.3.7]

Fix

  • target encoder in model.opt

[1.3.6]

ADD

  • add dosc on CV

[1.3.5]

Fix

  • Fix nans in targetencoder in CV

[1.3.4]

ADD

  • Target Encoding in CrossValidation
  • DenoisingAutoencoder in DataPrepare
  • Docs

[1.3.1]

Fix

  • Fix import - add loguru and psutil in requirements.txt

[1.2.28]

ADD

  • Advanced Logging (logs in .automl-alex_tmp/log.log)
  • Class Optimizer
  • Pruner in optimizer
  • connection with optuna-dashboard (run > optuna-dashboard sqlite:///db.sqlite3 )
  • NumericInteractionFeatures Class in data_prepare

[1.2.25]

Fix

  • Fix save & load in AutoML

ADD

  • Metod .score() and .fit_score() in Models
  • Class CrossValidation() examples in ./examples/03_Models.ipynb

[1.2.24]

Fix

  • same Fixses in AutoML

ADD

  • New info in Readme.md

[1.2.23]

A big update that changes the logic of work

NEW

  • Now processing the dataset is separated from the model for ease of use when you want to process the dataset yourself
  • Separate transform allows us to save and transfer processing to new data

ADD

  • Save & Load processing
  • Save & Load model
  • Reduce memory usage processing
  • Detect and remove outliers

[1.01.11]

Fix

  • score_cv_folds fix in ModelsReview
  • normalization

[0.11.24]

ADD

  • multivariate TPE sampler. This algorithm captures dependencies among hyperparameters better than the previous algorithm

Fix

  • "ValueError non-broadcastable output operand..." in AutoMLRegressor

[0.10.07]

Fix

  • DataConversionWarning in sklearn_models model.fit(X_train, y_train,)

[0.10.04]

Fix

  • verbose in LinearRegression

[0.08.05]

Fix

  • if y_train is not pd.DataFrame

[0.07.26]

Add

  • Calc predict policy in AutoML

Fix

  • timelemit in AutoML (deleted Catboost in optimization)

[0.07.25]

Add

  • Stacking in AutoML
  • fit on full X_Train (no_CV)
  • predict on full X in model_1 AutoML

[0.07.21]

Fixed

  • AutoML model_2 score

[0.07.20]

Add

  • Iterations in .opt

Fixed

  • timelemit in AutoML
  • Num Features Generator in empty Num Features list

[0.07.18]

Add

  • Features Generation in DataBunch
  • Features Selection in .opt
  • Generator interaction Num Features
  • Generator FrequencyEncoder Features
  • Generator Group Encoder Features
  • Normalization Data
  • Feature Importance

Fixed

  • RandomForest min_samples_split size
  • fix ModelsReview opt cv

[0.07.04]

Changed

  • remove target encoding
  • remove norm data
  • rebuild cross_val
  • preparation for the addition of FEs

[0.06.14]

Changed

  • add Docs in functions

Fixed

  • Try Fix .self buffer bug
  • Fix dataset size < 1000

[0.05.19]

Changed

  • Default stack_top=10 in AutoML

[0.05.16]

Changed

  • predicts in DataFrame

Added

  • predicts from configs

[0.05.11]

Added

  • RepeatedKFold in CV for prediction.
  • n_repeats=2 in .cv()

Changed

  • Stacking metamodel now LinearModel
  • in Stacking .predict n_repeats=2 => n_repeats=1 (timelimit :( )

Fixed

  • Fix Timelimit Error in Stacking