This attempts to fix the bug reported in issue #29 where the weights were contaminated with NaNs if there was missing data.
I took it on thinking it was an FHD issue, since I didn't look too closely at the repo, but hopefully my changes reflect the original code's intent.
This attempts to fix the bug reported in issue #29 where the weights were contaminated with NaNs if there was missing data. I took it on thinking it was an FHD issue, since I didn't look too closely at the repo, but hopefully my changes reflect the original code's intent.