lcrawlab / SINATRA

A statistical framework for feature selection and association mapping with 3D shapes

Home Page:https://projecteuclid.org/journals/annals-of-applied-statistics/volume-15/issue-2/A-statistical-pipeline-for-identifying-physical-features-that-differentiate-classes/10.1214/20-AOAS1430.full

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

MatrixtoSimplicialComplexTriangular - where?

muschellij2 opened this issue · comments

Not sure where MatrixtoSimplicialComplexTriangular is defined, but it's not in there.

Similarly - check_reconstruction

Good catch. The MatrixtoSimplicialComplexTriangular function is used for an older round of simulations where we used rbf-kernel interpolations, the current setting for generate_roc_curve_x is to specify mode='sphere'. The functions will be updated accordingly.

I'd have to get back to you on check_reconstruction @tsudijon can you help out here?

Thanks for raising this issue -- the check_reconstruction function is used in simulations which assess SINATRA's performance (using an ROC-type metric) and its sensitivity to parameters, in particular the functions roc_curve_simulation and metric_curve_simulation. The check_reconstruction function was used for an older use-case of these simulation functions, where the type parameter is given by cone. In all simulations reported, we use type=vertex which does not use the check_reconstruction function.

The type parameter controls how quality of the reconstruction is measured; we settled on type = vertex, which compares vertex-by-vertex the constructed region on a given shape to the known causal region on the same shape. The type=cone metric was something we experimented with earlier but was not as intuitive as a vertex-by-vertex comparison.

Potentially makes sense to remove the type='cone' functionality in these two simulation functions.

This can be resolved @lorinanthony !

Great!