transcriptomics

_sparse_nanmean makes two copies of the data matrix and performs a set index operation on a sparse array. It could be much faster by not doing this things.

Noticed while reviewing #1890.

from numba import njit, prange
import numpy as np

@njit(parallel=True)
def nanmean_lowlevel(data, indices, indptr, shape):
    N, M = s

The prose concat documentation could use an example of concatenating along the var axis.

Inspiration: #485