It can happen that, for a given threshold value, the PC scores cannot be computed because of a known issue in LAPACK(see here).
This was fixed in #95c2f20 by considering the threshold value as unstable and ignoring it, resulting in map_kfcv with an empty column corresponding to the threshold resulting in the error.
It could be nice to use a different matrix decomposition function (e.g. eigen instead of svd) that is more stable, if the error occurs.
It can happen that, for a given threshold value, the PC scores cannot be computed because of a known issue in
LAPACK(see here).This was fixed in #95c2f20 by considering the threshold value as unstable and ignoring it, resulting in
map_kfcvwith an empty column corresponding to the threshold resulting in the error.It could be nice to use a different matrix decomposition function (e.g. eigen instead of svd) that is more stable, if the error occurs.