icflow 0.0.2

A collection of simple utilities for machine learning workflows.

icflow is a Python package with some prototype 'workflow' tools for use in ICHEC.

Goals

The goal is to use it to help convert tools and approaches used by ML 'domain experts' into a more standardized workflow by:

• Analysing incoming Juyter notebooks, models, datasets and runtime environments (conda/container)
• Identifying 'hard-coded' data that can be moved into config files
• Adding utility scripts and methods for fetching data and models as needed
• Describing a study as a workflow, using some scripts here to 'stitch' the workflow together

Ultimately we will end up using some common workflow tools across ICHEC, likely something established and open-source (eg MLFlow) - this package is intended to understand and flesh-out our workflow needs and start transforming how we set up studies to ultimately move to these more standard tools.

Tests

In a Python virtual environment do:

pip install .'[test]'

Unit Tests

pytest

Linting and Static Analysis

black src test
mypy src test

All Tests

Requires tox:

tox