CrossPredict

• The library makes cross validation and reports generation easy
• Easy to extend to other models
• Supports Lightgbm, XGBoost, CatBoost
• Supports different crossvalidation strategies:
  • Supports crossvalidation by users (RepeatedKFold)
  • Supports stratified crossvalidation by target column (RepeatedStratifiedKFold)
  • Supports simple crossvalidation (RepeatedKFold)
• Easy use of target encoding with double crossvalidation
• Supports target encoding library category_encoders
• ML Pipeline building blocks

Table of contents

• Installation
• ReportsPreview
• Target Encoding with DoubleCrossValidation
• How to use
• Authors

Installation

python -m pip install crosspredict

ReportsPreview

#create report object
a = ReportBinary()

a.plot_report(
    df,
    report_shape = (5,2),
    report={'Roc-Auc':  {'loc':(0, 0)},
          'Precision-Recall': [{'loc':(0, 1)}],
          'MeanTarget-by-Probability': [{'loc':(1, 0)},{'loc':(1, 1)}],
          'Gini-by-Generations': {'loc':(2,0), 'colspan':2},
          'MeanTarget-by-Generations': {'loc':(3,0), 'colspan':2},
          'Probability-Distribution': [{'loc':(4,0)},{'loc':(4,1)}]},
    cols_score = ['result_egr_to_one','probability'],
    cols_target = ['target','target2'],
    col_generation_deals='first_dt_no_comm_mon'
)
a.fig.savefig('report1.png')

a.plot_report(report_shape = (3,2),
              report={'roc-auc':  {'loc':(0, 0)},
                      'precision-recall': {'loc':(0, 1)},
                      'mean-prob': {'loc':(1, 0)},
                      'gen-gini': [{'loc':(2,0), 'colspan':2}],
                      'distribution': {'loc':(1,1)}
                     },
              cols_score=['probability']) 
a.fig.savefig('report2.png')

Target Encoding with DoubleCrossValidation

# creates iterator
iter_df = Iterator(n_repeats=3,
                    n_splits=10,
                    random_state = 0,
                    col_client = 'userid',
                    cv_byclient=True)

# fits target encoder (creates mappings for each fold)
cross_encoder = CrossTargetEncoder(iterator = iter_df,
                                    encoder_class=WOEEncoder,
                                    n_splits= 5,
                                    n_repeats= 3,
                                    random_state= 0,
                                    col_client= 'userid',
                                    cv_byclient= True,
                                    col_encoded= 'goal1',
                                    cols= ['field3','field2','field11','field23','field18','field20']
                                  )
cross_encoder.fit(train)

# train cross validation models
model_class = CrossLightgbmModel(iterator=iter_df, 
                                 feature_name=feature_name,
                                 params=params,
                                 cols_cat = ['field3', 'field2', 'field11', 'field23', 'field18', 'field20'],
                                 num_boost_round = 9999,
                                 early_stopping_rounds = 50,
                                 valid = True,
                                 random_state = 0,
                                 col_target = 'goal1',
                                 cross_target_encoder = cross_encoder)
result = model_class.fit(train)

How to use

ML Pipeline Calculation - ML_Pipeline.ipynb
Configuration file for pipeline - params.yaml

Plot_Reports_for_Binary_Classification_problem_example - Plot_Reports_for_Binary_Classification_problem_example.ipynb

Simple_example_in_one_Notebook - Simple_example_in_one_Notebook.ipynb

Iterator_class - Iterator_class.ipynb

CrossModelFabric_class - CrossModelFabric_class.ipynb

CrossTargetEncoder_class - CrossTargetEncoder_class.ipynb

Authors

Vladislav Boyadzhi Alexander Tarelkin