Notes on pb's notes

[sophiacofone]

• Makefile: jesse fixed, right?
• requirements.txt: jesse fixed, right?
• Q: Did you normalize all landmarks for all images before analysis? : we should be extra clear when we are using normalized landmarks and when we aren't (personally I think we did this already but could be wrong)
• Q: Did you try to use landmarks in the analysis before normalization?: again should be clear when we are doing this
• Q: Did your normalization reproduce results/preprocessing in Michael Wan's paper?: not sure what this means, does he mean reinforcing the findings from Table II?
• svc.md: add EDA, fix the screenshots of the text output, fix labels of 0 and 1, generally needs more documentation/clarity on the scatterplots, reproducibility (add your make targets)
• Seems logreg uses data from here: https://github.com/ostadabbas/Infant-Facial-Landmark-Detection-and-Tracking If so, then why is preprocessing.md focussed so much on "raw", which would have all sorts of potential problems?: I don't really understand this comment
• sc_notes_logreg.md -- the link to derived.md is broken: I fixed this
• bayes.md: provide clarity on the UN-normalized landmarks

[jhautala]

I fixed the Makefile but I couldn't reproduce the error with installing requirements from requirements.txt (I created a new conda environment with python 3.10, used pip install -r to install dependencies, and was able to execute make targets). If anyone can dupe his error, I can help troubleshoot, but maybe we should just make it extra clear that file is for reference only.

[sophiacofone]

hmm ok! thanks for testing that!

[jhautala]

I created a conda env file and tested to make sure it works (should address some concern about environment reproducibility).

[ZongyuWu97]

svc.md fixed

[sophiacofone]

I added additional testing for un normalized vs normalized for log reg and updated the notes: https://github.com/ds5110/faces/pull/36