ML Template

ML template is an easy-to-use tool to automate the boilerplate code for most machine learning projects.

This tool creates a user-oriented project architecture for machine learning projects.

Modify the code under #TODO comments in the template project repository to easily adapt the template to your use-case.

How to use it?

1. Install the package as - pip install mltemplate
2. Then, simply run mltempate <project-name> from your terminal to create your template

And Voila!

This creates a project directory in your current folder similar to -

template
├── Dockerfile.cpu
├── Dockerfile.gpu
├── LICENSE.md
├── Makefile
├── README.md
├── jupyter.sh
├── requirements.txt
└── template
    ├── __init__.py
    ├── __main__.py
    ├── cli
    │   ├── __init__.py
    │   ├── predict.py
    │   └── train.py
    ├── notebooks
    └── src
        ├── __init__.py
        ├── models.py
        ├── datasets.py
        └── transforms.py

All you have to do next is -

1. Update python frameworks and versions in template/requirements.txt as need for your project
2. Head to template/datasets.py and modify create a new dataset that will work for your use case
3. Navigate to template/models.py and create a new model class with your sota (or not) architecture
4. In template/transforms.py add transforms such as Normalizer, Denormalize etc.
5. Follow the TODO steps in template/cli/train.py and template/cli/predict.py to make the necessary changes

Checkout the README.md in the template directory for further instructions on how to train, predict and also monitor your loss plots using tensor board.

Future Work

Currently, this package only supports boilerplate creation for ML projects in pytorch

We plan to support keras in the future.

Development

Local Testing

Run the following command to generate the packaged library

poetry build

Install the library from the generated whl file using

pip3 install dist/mltemplate-<version>-py3-none-any.whl --force-reinstall

You can then test the functionality of the pypi package.

Publish package to PyPi

To create a new version of the framework, update the version in pyproject.toml file Merge your changes to main and then publish git tag to trigger the release ci

git tag <x.x.x>
git push origin <x.x.x>

License

Copyright © 2020 Sowmya Yellapragada

Distributed under the MIT License (MIT).