memory requirement for large-ish dataset

[yanlingpan]

I tried to run PHATE on a dataset with 120k observations and 28,426 variables, and ran out of memory (at the step "Calculating affinities") with 490G CPU memory using phate.PHATE(mds_solver='smacof'). All other parameters were default.

Because my dataset is much smaller than the 1.3 million mouse brain cell dataset from 10x, as tested in the paper, I am wondering what the memory requirement is to run the 1.3 million cell dataset in 2.5 h with 2000 landmarks? Was this 1.3 million cell dataset run with all default parameters? Are there any other parameters I should adjust?

Thank you.