nasa-impact / bps-numerical Goto Github PK

Attached is an SVG that entails tentative pipeline/components used for the overall process:

Basically, the pipeline consists of 4 significant steps:

• data pre-processing
• training (either single phenotype or multiple phenotypes)
• feature scoring (uses the trained model to figure out the common "important" features across multiple phenotypes)
• feature ranker (this is a WIP where we train several independent models for a single phenotype and get relevant features for that phenotype)

In case, anyone wants to explore the excalidraw diagram:
https://excalidraw.com/#json=5K9u7CzNo-atbuJdDM6RH,PfZkw_iiEvHdPDxm0lBeBQ

cc: @xhagrg @code-geek @muthukumaranR

Here are the things remaining to wrap-up the project:

• Consolidate multiple branches/features into a single (and PR)
• Add MRR-based module to official code (previously only on jupyter notebook)
• Bugfix a few things that are only done in the experimental notebook
• Create a final notebook pipeline that anyone can use
• Update the paper manuscript with final results

cc: @xhagrg @muthukumaranR