PN-ML-Challenge

The model parameters used in pipeline can be adjusted by changing .env file.

Variable	Description	Example Value
TEST_SIZE	How big the test size should be	0.3
MAX_R2	Max allowed R^2	0.03

In case model will exceed max allowed R^2 then script: classifier/stages/evaluate_model.py will log a warning.

---

Features

After installing virtual environment make install
activate: . .venv/bin/activate and make help to see help message.

Please use make target where target is one of:
all                 commit and push all changes
brew-allure         install allure with brew long
bump                (PART= ) bump the release version - deduced automatically from commit messages unless PART is provided
changelog           (UNRELEASED= current version) update the changelog incrementally.
check-commit        check the commit message is valid
clean               clean up temp and trash files
commit              make interactive conventional commit
docs                render documentation
format              format code
get-version         output the current version
help                display this help message
install             install the requirements
lint                run static code checkers
pre-install         install pre-requirements
release             create a new github release
tag                 pull tags and tag a new version
test                (ALLURE=True BROWSE=True) run tests
venv                install virtual environment

Interactive commits cooperating with pre-hook and changelog updates:

---

Continuous Integration and delivery

Complete CI-CD process will be visible in GitHub Actions page containing:

• CI - run on every push to master and on pull requests:
  • Test - run unit tests.
  • Format - check if code is formatted.
  • Lint - run all static code checkers with prospector.
• CD - continuous package delivery:
  • Preprocessing data - load data from origin and perform basic operations.
  • Feature engineering - create features and split dataset.
  • Train model - train model with parameters from .env and save classifier/models/ directory.
  • Evaluate model - score model and save output to classifier/scores directory.
• CI/CD - manual trigger, includes CI and CD pipelines connected.

---

New version of model creation

Pipeline will always use a model with version specified in VERSION file.
In case of making changes to a model update the recent version:

• make bump - will bump the version file of a model based on last commit created with commitizen
• make changelog UNRELEASED='cat VERSION' - update changelog with recent commit history
• make all - commit all files to repository

When a new version of a model will be created stages scripts will create new model file:
classifier/data/models/finalized_model-v0.1.0.sav
and a new score file:
classifier/data/scores/model-v0.1.0.txt
This value can be used to determine if a new model should be deployed or rollbacked.