search

neurosutton / connectivity

2 stars 0 forks source link

nia_stats_and_summaries.calculate_auc(): specifying subnetwork measure does not work #13

Closed joshbear closed 3 years ago

joshbear commented 3 years ago

The code will run when you call calculate_auc() with a df and a subgroup measure, but the output is nonsensical. This could be related to the other issue raised about separately analyzing each subnetwork.

CODE: nss.calculate_auc(df, msrs=['sg_num_connected_comp'])

OUTPUT: 1%| | 28/4999 [00:00<00:17, 276.81it/s] Working through ['sg_num_connected_comp'] for FCN: None SG_NUM_CONNECTED_COMP eses - hc = nan 100%|██████████| 4999/4999 [00:16<00:00, 310.68it/s] The experimental AUC difference, nan, occurs 0.0% of the time in the boostrapped results.

image
neurosutton commented 3 years ago

I think #10 , #12 , and this one are all related. The FCN "none", which is whole brain, does not have subgraph metrics associated with it. I will put in a check to make sure that comparisons are not made for metrics that do not exist.

neurosutton commented 3 years ago

10, #12, and this one were all related to (1) the 'nan' string issue and/or (2) a subtle change adding a recorded threshold to subgraph analysis that must have been committed after you started processing the graphs. I'll close all three issues. Please reopen, if you are still having trouble.