ML model inference support to be stripped from dclab

DC-analysis / dclab

Python library for the post-measurement analysis of real-time deformability cytometry (RT-DC) data sets

https://dclab.readthedocs.io

Other

10 stars 12 forks source link

ML model inference support to be stripped from dclab #236

Closed paulmueller closed 2 days ago

paulmueller commented 11 months ago

Since ML model inference via ancillary features is not used AFAIK, and it has become a burden to maintain (failing tests with tensorflow, long pipeline times), I am planning to strip this functionality from dclab. This also affects the corresponding functionality in Shape-Out (loading models).

This will not affect the presence of the ml_score_??? features and I might keep in the registration of features (where feature names can be mapped to proper axes labels).

Please comment here if you have an opinion.

paulmueller commented 10 months ago

TODO:

[ ] Replace Python "3.11" with "3.x" in GHA (tensorflow always lacked behind)

B-Hartmann commented 2 months ago

If it is not used, remove it to keep the code base smaller and easier to maintain.

The inference would work on-the-fly and dclab would not save the inference information to the file, is that correct? If yes, I think most people would anyway do the inference separately and create a new .rtdc file which they would then inspect the data with Shape-Out. On-the-fly inference is too time and resource consuming to do it every single time again.