jovo / signal-subgraph-classifier Goto Github PK

run simulation again to see whether s.hat=1 is reliably good. also, make ylims bigger to see errorbars.

twould be nice if i had a folder called fig_scripts, in which each script generated a single figure.
to do this, i'd need to make some relatively simple modifications,
but navigating between figures and code and tex would be much simpler.
each figure name would be: [fname (param) (value) (param) (value) ....]
i think definitely worth doing, in general.

tex should state the script that generated the figure.

how do we choose which one upon having equally valued ones?

running things long enough should yield smoother plots

easy: the max pvalue of the edges in the signal subgraph tells us our confidence. lower is better, of course.

when we have data from multiple different centers, we can deal with that appropriately.

although we can threshold at each p-value, incrementing the threshold often means many edges now satisfy the constraints. in general, we have no way of choosing a subset of edges with equal weights, so we should just choose all of them. this should speed up the search dramatically, and remove some unnecessary noise in the results.

two versions (one for edges, one for vertices) of

|Shat \cap S| / |Shat \cup S|?

a roc curve for each?

x-axis is # of training samples

for resubmission perhaps.