1_LearningFE

Instructions for running LearningFE Training for Recsys2023

1. Introduction

We prepare train data with variety of feature engineering to catogorical features and numerical features, which includes GlobalCountEncoding, NewValueOneHotEncoding and Indexing.

Feature engineer pipeline overview. Role1 and Role2 refer to users and ads in this dataset. The same holds for rest 'Role' terms in this solution.

After done feature engineeirng, processed train data will be used in two manners.

• Training with LGBM to do inference on test data
• Combining with GNN embeddings and training with LGBM/XGBoost to do inference on test data. (How to generate GNN embeddings is decribed in 0_train/2_supGNN, 0_train/3_BiGNN and 0_train/4_simGNN)

2. Getting start - Train data

# 0. We have assumption that train data has been located under data/sharechat_recsys2023_data/train

step1:
    open 0_data_prepare.ipynb and run through whole script

step2:
    open 1_train_LGBM.ipynb and run through whole script

3. Go to other folders under 0_train to create GNN embeddings

• 0_train/2_supGNN
• 0_train/3_BiGNN
• 0_train/4_simGNN

4. Go for inference

please switch to 1_inference/1_LearningFE