search

deepskies / DeepDiagnostics

Inference diagnostics for mostly SBI
MIT License
1 stars 0 forks source link

Plotting and Metric for Local classifier 2 sample test #63

Closed voetberg closed 6 months ago

voetberg commented 6 months ago

Closes #39 Following the paper from here: https://arxiv.org/abs/2306.03580 and discussion here: https://github.com/deepskies/OpticalClusterSBI/discussions/3

Produces plots that look like this:

image

image

bnord commented 6 months ago

what's the difference between the top and bottom figures? is the bottom one a 2d histogram of something? What's the yellow?

voetberg commented 6 months ago

Difference between them is the the dimension of the histogram yeah (https://github.com/JuliaLinhart/lc2st/blob/main/lc2st/graphical_diagnostics.py#L167 vs https://github.com/JuliaLinhart/lc2st/blob/e221cc326480cb0daadfd2ba50df4eefd374793b/lc2st/graphical_diagnostics.py#L187 ). We can add a colorbar in but the colors are just a higher probability of being corrected predicted for that region (as defined by the local classifier).

voetberg commented 6 months ago

The best way to think of the histogram here is that it's a single part of a corner plot

bnord commented 6 months ago

the legend just applies to the cdf plot? if so, do you know what the yellow is in the top 1-d hist?

voetberg commented 6 months ago

Colorcoding is the same, it's just the regions of the parameter that have a higher probability of being correctly classified

voetberg commented 6 months ago

Legend is for the cdf, yes.