Bad clasification performances : sometimes multitask_sfan.py does not return any selected features

[nanls]

Output of code/multitask_sfan.py

Commented lines give the parameter values (2 for sfan, 3 for msfannp and msfan).
Then, each line corresponds to one tasks, and is a space-separated list of node indices, starting at 1.

Problem

Sometimes, for msfan, these lists are empty for some folds only, while for other folds, the selection was pretty good.
This heterogeneity causes poor performances, with hight variance. Moreover, no phenotype predictions can be made.

These observations have been made with a small number of SNPs (200 and 1000).

Line of approach

Try with bigger number of SNPs
Try with another choice of hyperparams grid