Add support for proportions and softmax normalisation - Githubissues

thenineteen / Semiology-Visualisation-Tool

Data driven 3D brain visualisation of semiology. Semiology to anatomy translator based on over 4600 patients from 309 peer-reviewed articles.

MIT License

9 stars 6 forks source link

Add support for proportions and softmax normalisation #201

Closed fepegar closed 3 years ago

fepegar commented 3 years ago

Resolves #8 for now.

thenineteen commented 3 years ago

Thanks this is great! As per previous discussions in JD meetings and with data guys, I've implemented a proportions options #118 #8 . This would still be backward compatible using either of your other two options "Rescaling" and the other new options "Softmax".

The combination of proportions is simply their mean, and no further rescaling or softmax or other normalisation is required. A weight for each proportion has not been envisaged because this would treat the semiologies as unequal based on number of datapoints which exacerabted the publication bias towards certain semiologies (in the rescaling and softmax versions we also do not implement any sort of weighting either)

thenineteen commented 3 years ago

Also resolves #118 (NB semiology level proportions, not study level)

thenineteen commented 3 years ago

doesn't let me request a review from the initiator of the pull request? @fepegar all the changes are just to incorporate an extra option to the minmax and softmax options added here, namely the proportions option. Some changes are because no scaling is performed for proportions when combining.

thenineteen commented 3 years ago

has an issue with now only updating colours for combinations of semiologies on 3D slicer but worked in VS code. fixed in commit below.

thenineteen commented 3 years ago

@fepegar just with the proportions option, can you please make the table show to one decimal place? e.g. in below, the left hippocampus should show 4.0% (3.97%)

It will then be ready to squash and merge