Repository navigation
Flexible MLflow workflow for lab/external users #1348
Description
I'll bring it up in the dev meeting today, just leaving a note:
I think MLflow integration at the HSSM level can make sense, but we probably have to reflect on the actual use-case a little there.
- Linking HSSM runs to the network origin lineage, makes sense for initial proofing of bringing a network into production.
- One can imagine someone downstream (independent researcher) wanting to track their own numerical experiments, or even their literal overall data analysis workflows. In that case, the linkage to our oscar level file lineage might not be too useful.
I am concerning with making things useful for the latter case.
Originally posted by @AlexanderFengler in #1320 (comment)
Activity
linear-code commentedon Sep 24, 2026
HSSM-849
cpaniaguam commentedon Sep 25, 2026
@AlexanderFengler I think the framework introduced in #1320 is flexible enough to support both use cases with some adjustments in defaults and documentation.
The scope I have in mind: make tracking flexible enough that someone working only in HSSM can record what they did and find it again.
Anything more ambitious, such as recording model comparisons, computing comparison metrics automatically, we would rather leave until the basics have seen some use.
Comments welcome, particularly if this is narrower than you had in mind.