kaggle-otto2

Our 20th place solution of the Kaggle OTTO – Multi-Objective Recommender System competition

Code Structure

Solution Summary

Solution details are written in the Kaggle Discussion: https://www.kaggle.com/competitions/otto-recommender-system/discussion/382771

3 Types of experiment environment

My code has 3 expriment environemt

1. dev: For faster local experiment. With 1/20 sampled cv data (e.g. ./yaml/exp001_dev.yaml)
2. cv: For local experiment and validation (e.g. ./yaml/exp001_cv.yaml)
3. lb: For submission (e.g. ./yaml/exp001_lb.yaml)

Procedure

Please see ./demo.ipynb for the execution log example

000. Setup poetry

# Install packages
poetry install

# Get into virtual env
poetry shell

000. Download datasets

# Download datasets
# ※ ~/kaggle/.kaggle.json with your Kaggle API Key is required
./bin/000_download.sh

001. Preprocess

./bin/001_preprocess.sh exp001_dev

002. Candidate Generation

./bin/002_candidate_generation.sh exp001_dev

003. Feature Engineering

./bin/003_feature_engineering.sh exp001_dev

004. Train Ranker

./bin/004_train_ranker.sh exp001_dev lgbm