Currently cusignal is needed for argrelmax. cusignal is a pain to install and incompatible with pip, you need to use conda.
cusignal also does some weird precompile at installation time, so you can't just grab the code from the repo and throw it in to hyperseti codebase.
Best approach will be to carefully write our own peak finding algorithm on the GPU. This is currently the bottleneck of the whole pipeline so most fruitful area to optimize.
Currently cusignal is needed for
argrelmax
. cusignal is a pain to install and incompatible withpip
, you need to useconda
.cusignal also does some weird precompile at installation time, so you can't just grab the code from the repo and throw it in to hyperseti codebase.
Best approach will be to carefully write our own peak finding algorithm on the GPU. This is currently the bottleneck of the whole pipeline so most fruitful area to optimize.