Ayniy, All You Need is YAML

Ayniy is a supporting tool for machine learning competitions.

Documentation | Slide (Japanese)

# Import packages
from sklearn.model_selection import StratifiedKFold
import yaml

from ayniy.model.runner import Runner

# Load configs
f = open('configs/run000.yml', 'r+')
configs = yaml.load(f, Loader=yaml.SafeLoader)

# Difine CV strategy as you like
cv = StratifiedKFold(n_splits=5, shuffle=True, random_state=7)

# Modeling
runner = Runner(configs, cv)
runner.run_train_cv()
runner.run_predict_cv()
runner.submission()

Examples

Platform	Competition Name	Rank	Repository
CodaLab	Basketball Behavior Challenge BBC2020	1	Link
Nishika	財務・非財務情報を活用した株主価値予測	2	Link
SIGIR2021	SIGIR eCOM 2021 Data Challenge	3	Link
SIGNATE	ひろしまQuest2020#stayhome【アイデア部門】	-	Link
ProbSpace	YouTube動画視聴回数予測	6	Link
Solafune	夜間光データから土地価格を予測	6	Link
atmaCup	#8 [初心者向] atmaCup	-	Link
Kaggle	WiDS Datathon 2021	64	Link
Kaggle	Titanic: Machine Learning from Disaster	-	Link

Starter Kit

Scripts

mkdir project_dir
cd project_dir
sh start.sh

kaggle_utils is used for feature engineering.

Environment

docker-compose -d --build
docker exec -it ayniy-test bash

MLflow

cd experiments
mlflow ui -h 0.0.0.0

Kaggle Notebook

!git clone https://github.com/upura/ayniy
import sys
sys.path.append("/kaggle/working/ayniy")
!pip install -r /kaggle/working/ayniy/requirements.txt
!mkdir '../output/'
!mkdir '../output/logs'
from sklearn.model_selection import StratifiedKFold
from ayniy.model.runner import Runner

For Developers

Test

pysen run lint
pysen run format

Docs

In container,

cd docs
make html

Out of container,

sh deploy.sh

https://github.com/upura/ayniy-docs