Key repositories
composable-logs/mnist-digits-demo-pipeline
- Repo for the mnist demo training pipeline.
- A public task and experiment tracker for this pipeline is hosted on Github Pages:
composable-logs/composable-logs (Python)
- Repo for the main library (Python, Ray, OpenTelemetry). This is used when running the demo pipeline.
- Dependency for the above demo pipeline.
- Install via PyPI:
composable-logs/mlflow (Javascript)
- A slightly modified version of MLFlow, that can build static MLFlow-like sites where all metadata is included in the front end.
- Dependency for the above demo pipeline.
- The static assets are made available as a Python package via PyPI:
- Install as
pip install composable-logs-webui --target <install dir>
(latest release, PyPI link). After installation the assets will be in theassets
directory.
- Install as
Development environments are dockerized, and makefile:s should be available for most common tasks.
Inspecting the Github actions in the composable-logs
and mnist-digits-demo-pipeline
repos can also be useful since they show how various tests (unit, static code analysis) and pipelines are run in docker.
Instructions assume linux-based setup (eg. Ubuntu), but should also work on macs with some changes (or at least on non-M1 ones).