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