Readme

Description

This repo was generated for participating in the upstage contest. I got 0.8598 AUC, conducting ensemble catboost and lgb model from tabular data.

• Goal : Binary classification
  Using log data(2009/12~2011/12) from 5914 users, predict the probability of each customer's total purchases exceeding 300 on December 2011.

• Module execution : Inference > feature engineering > feature generation

  • inference.py
    With parsed arguments, print out Out Of Fold Validation Score by using StratifiedKFold from sklearn.model_selection.
  • feature_generation.py
    By adding or subtracting required columns, manipulate pandas dataframe.
  • feature_engineering
    • Execute functions : generate label, feature preprocess, and feature generation
    • By using the aggregation function, grow features or merge features

Installation

pip install -r requirements.txt at ./code dir.

catboost==0.24.4
lightgbm==3.1.1
matplotlib==3.1.3
numpy==1.19.5
pandas==1.1.5
scikit-learn==0.24.1
seaborn==0.11.1
xgboost==1.3.3

Example

inference.py

From arg parser, you can choose one from various options.

parser.add_argument('--seed', type=int, default=0, help="base seed is 0")
parser.add_argument('--ym', type=str, default='2011-12', help="add target year_month to predict, base is 2011-12")
parser.add_argument('--engineering', type=str, default='feature_engineering_all', help="choose feature engineering type")
parser.add_argument('--ensemble', type=bool, default=True, help="choose true if you want to ensemble model, else choose false")

train, test, y, features = getattr(import_module("feature_engineering"), args.engineering)(data, year_month)

if args.ensemble:
    oof_xgb, xgb_pred, fi = make_xgb_oof_prediction(train, y, test, features, model_params=xgb_params)
    oof_lgb, lgb_pred, fi = make_lgb_oof_prediction(train, y, test, features, model_params=lgb_params)
    oof_cat, cat_pred, fi = make_cat_oof_prediction(train, y, test, features, model_params=cat_params)
    ...

Contraints

• Input directory doesn't exist!
  • train.csv and sample_submission.csv was not uploaded. But the column info was opened below.
• Columns of train data

Improvements

• Things to improve.
• Add Feature selection (sample)
• Using Tabnet architecture. (sample1, sample2)