ML project template

This repository is a template directory for ML project and inspired by upura.

Structures

.
├── configs
│   └── config.json(not included here)
├── data
│   ├── input
│   │   ├── sample_submission.csv
│   │   ├── train.csv
│   │   └── test.csv
│   └── output
├── docker
│   └── Dockerfile
├── features
├── logs
├── models
├── notebooks
├── src
├── utils
│   ├── data_loader.py
│   ├── feature_base.py
│   ├── feature_create.py
│   └── convert_to_feather.py
├── .gitignore
├── LICENSE
├── README.md
└── run.py

Directory Explanation

• config:
  • model: model parameters
  • config.yaml: ML settings
• data
  • input: contains original data or feather files.
  • output: contains csv file for submission.
• docker: contains Dockerfile and docker-compose.yml
• features: contains features created by train and test data.
  • importance: feature importances
• fig: contains some figures.
• logs: contains logging data including features, a model, parameter and cv scores.
• models: contains saved model.
• notebooks: contains EDA codes.
• src: contains model source codes and project-specific useful codes.
• utils: contains generally useful codes.
• requirements.txt

setup

• docker-compose up -d: prepare docker container
• docker-compose run python bash: start bash

Actions

• cd utils && python convert_to_feather.py: Convert csv files to feather files.

• python feature_create.py: Create features in feather files.

• cd .. && cd src && python run.py: Start learning.