manage many candidates in a segment
caseyjlaw opened this issue · comments
Bad data can trigger candidates in every image. This is slow and probably could lead to memory problems.
Look for ways to avoid being saturated by candidates.
Current design makes it hard to do this and maintain parallelism between fftw and cuda functions. Postponing for now, as this problem (too many candidates) should be controlled by cleaning data first.
Clustering implemented via hdbscan. Peak SNR of each cluster is actually plotted, etc.