Release Note

Version 0.2.0

This release add following new features:

Feature engineering

  • Feature generation

  • Feature selection

Data clean

  • Special empty value handing

  • Correct data type

  • Id-ness features cleanup

  • Duplicate features cleanup

  • Empty label rows cleanup

  • Illegal values replacement

  • Constant features cleanup

  • Collinearity features cleanup

Data set split

  • Adversarial validation

Modeling algorithms

  • XGBoost

  • Catboost

  • LightGBM

  • HistGridientBoosting

Training

  • Task inference

  • Command-line tools

Evaluation strategies:

  • Cross-validation

  • Train-Validation-Holdout

Search strategies

  • Monte Carlo Tree Search

  • Evolution

  • Random search

Imbalance data

  • Class Weight

  • Under-Samping - Near miss - Tomeks links - Random

  • Over-Samping - SMOTE - ADASYN - Random

Early stopping strategies

  • max_no_improvement_trials

  • time_limit

  • expected_reward

Advance features:

  • Two stage search - Pseudo label - Feature selection

  • Concept drift handling

  • Ensemble