BanditPAM
BanditPAM copied to clipboard
Precomputed loss
A workaround for #163 would be if I could simply provide a precomputed loss matrix, and use "precomputed" loss.
I couldn't find the doxygen documentation online anywhere, the available losses appear to be here:
https://github.com/ThrunGroup/BanditPAM/blob/5f12c07d80101abf076eea32a313bba9dbf427c8/src/kmedoids_algorithm.cpp#L129
I like the proposed solution, as supporting sparse matrices natively would be significantly more complicated
It does increase the run time with the quadratic cost of the similarity matrix though.