Closed quattro closed 1 year ago
This is more or less done, but limited to the BCOO
datatype in JaxSparse. Leaving this open for now, if we can extend to BCSR
at some point.
Another enhancement would be a "blocked" or "batched" version of this that captures different genotype matrices over different chromosomes. This would have a benefit for downstream solvers by treating the large covariance matrix as block diagonal.
Catch-all for implementing centered/standardized genotypes as a sparse linear operator