Support MLFlow as an experiment tracker
Opened this issue · 0 comments
dmcc commented
This is a somewhat big change -- while we tried to minimize dependencies on Weights & Biases, there are certainly many. This bug will start with a growing list of known dependencies:
- Artifacts are based entirely on W&B's model (even down to naming restrictions). This would need to be factored out.
- Documentation: Much (example) is W&B specific.
- Aeroview also assumes W&B artifacts.
Specifying which backend to use would likely be specified in Aeromancy configuration in pyproject.toml
.