Testing a Machine Learning pipeline

This project showcases in practise how tests are written and what are they in general when talking about software development.

All the project is heavily influenced by the great book "Unit Testing. Principles, Practices and Patterns" by Vladimir Khorikov. You can find the book here.

All the tests are in the tests directory.

The framework used to run the tests is pytest.

Python environment

We will use virtualenv for managing virtual environments. To install it, run:

# Get the latest version
pip install virtualenv

To create an empty Python 3.9 environment which is used in this project run the command:

virtualenv --python 3.9 app_env

To activate the virtual environment, run the following command:

source app_env/bin/activate

All the packages with the fixed versions are in the requirements.txt file. To install the packages, run:

pip install -r requirements.txt

Article on testing

The main article that supplement this code can be accessed here

This article is written by me and covers most of the basic topics in testing that will be needed to write robust software.

Application in question

The application is a machine learning pipeline that trains a classifier:

f(text) -> class

The text is a customer review, and the class is the overall score of the product.

The pipeline creates a model where given a text, the ML model predicts the review score of the product (1 - 5 stars).

To run the whole pipeline use the command:

python -m pipeline.pipeline

Running the tests

The tests are grouped into tests, integration and end to end tests:

├── end_to_end
│   ├── data
│   ├── __init__.py
│   └── test_end_to_end.py
├── __init__.py
├── integration
│   ├── data
│   ├── __init__.py
│   ├── test_creating_fitting.py
│   └── test_reading_cleaning.py
└── unit
    ├── data
    ├── __init__.py
    ├── test_clean_data.py
    ├── test_evaluate_model.py
    ├── test_model_fitting.py
    ├── test_model_input_preparation.py
    └── test_read_data.py

To run all the tests, use the command:

pytest

With coverage report:

coverage run -m pytest && coverage report