Using the fashion-mnist dataset to demonstrate ML-pipelines and experimentation with DVC.
DVC studio shows you a visualization of the ML-project lifecycle.
To build a docker image:
docker build -t fashion-mnist --build-arg USER_ID=$(id -u) --build-arg GROUP_ID=$(id -g) -f Dockerfile .
To start a docker container, expose GPU (optional), mount current directory and connect to the container:
docker run -d --rm -it --gpus all --volume $(pwd):/workspace --name fashion-mnist fashion-mnist
docker exec -it fashion-mnist /bin/bash
Use doit to run linting and tests:
doit lint
doit pytest
Use dvc to create reproducible ML-pipelines and experiments with git tracking:
dvc repro
dvc exp run
Use mlflow to run lots of experiments that you do not want to track with git.
├── LICENSE
├── dodo.py <- Makefile-like multiplatform CLI
├── README.md <- The top-level README for developers using this project.
├── data
│ ├── external <- Data from third party sources (ex. script config files)
│ ├── interim <- Intermediate data that has been transformed.
│ ├── processed <- The final, canonical data sets for modeling.
│ └── raw <- The original, immutable data dump.
│
├── docs <- Documentation
│
├── models <- Trained and serialized models, model predictions, or model summaries
│
├── notebooks <- Jupyter notebooks. Naming convention is a number (for ordering),
│ the creator's initials, and a short `-` delimited description, e.g.
│ `1.0-jqp-initial-data-exploration`.
│
├── references <- Data dictionaries, manuals, and all other explanatory materials.
│
├── reports <- Generated analysis as HTML, PDF, LaTeX, etc.
│ └── figures <- Generated graphics and figures to be used in reporting
│
├── requirements.txt <- The requirements file for reproducing the analysis environment, e.g.
│ generated with `pip freeze > requirements.txt`
│
├── setup.py <- makes project pip installable (pip install -e .) so fashion_mnist can be imported
├── fashion_mnist <- Source code for use in this project.
│ ├── __init__.py <- Makes fashion_mnist a Python module
│ │
│ ├── data <- Scripts to download or generate data
│ │
│ ├── features <- Scripts to turn raw data into features for modeling
│ │
│ ├── models <- Scripts to train models and then use trained models to make
│ │ predictions
│ │
│ └── visualization <- Scripts to create exploratory and results oriented visualizations
|
├── Dockerfile <- Dockerfile with settings to run scripts in Docker container
├── dvc.yaml <- DVC pipeline; see dvc.org
├── params.yaml <- Parameter values (things like hyperparameters) used by DVC pipeline
├── setup.cfg <- config file with settings for running pylint, flake8 and bandit
└── pytest.ini <- config file with settings for running pytest
Project based on the cookiecutter data science project template. #cookiecutterdatascience.