This repository contains a set of example projects for mllint, the command-line utility to evaluate the quality of Machine Learning (ML) projects by means of static analysis of the project's repository.
Each branch of this repository, except main, implements a ML project written in Python that has used mllint to generate a project quality report (see report.md)
Note: it is usually not recommended to commit
mllintreports to the project repository, similar to how code coverage reports should also not be committed to the repository. This is only done in this repository for demo purposes.
The branches 1-*-basic* implement a basic ML project, that starts out with a requirements.txt to track Python dependencies, no code or data quality linting and no data versioning or CI. mllint therefore also gives this version of the project very poor scores. Over the course of several iterations, we implement several improvements in the project, guided by mllint's recommendations. A description of these improvements, along with the branch of this repository that stores this iteration, can be found in the table below.
| # | Branch | Improvements relative to previous iteration |
|---|---|---|
| 1 | 1-1-basic |
- |
| 2 | 1-2-basic-dvc |
Use Data Version Control (DVC) to track data dependencies, instead of downloading from Drive folder with a data/get_data.py script. |
| 3 | 1-3-basic-poetry |
Start using Poetry |
| 4 | 1-4-basic-linting |
Add Pylint, Mypy, Black, isort and Bandit to dev dependencies, create configurations for Pylint and isort |
| 5 | 1-5-basic-fix-codestyle |
Run black . and isort . to fix code style issues |
| 6 | 1-6-basic-fix-pylint |
Fix all Pylint's code quality issues |
| 7 | 1-7-basic-fix-bandit |
Fix all Bandit's security issues |
| 8 | 1-8-basic-fix-mypy |
Fix all Mypy's type-checking issues. |
| 9 | 1-9-basic-fix-ci |
Implement a basic CI pipeline to install the project's dependencies and run mllint on it, saving the report it generates as a pipeline artifact (e.g. this artifact). |
Based on the DVC Get Started project from the creators of Data Version Control (DVC)