fastppm is also extreme easy to use! checkout our github repository for instructions:

@t2mykleb thanks, this is helpful. i also suspect that for low-precision, we are better off using EM. since the answer is model specific, however, it may be best (and informative) to simply test both approaches.

@sp_monte_carlo yes, i am trying to tease out how much is cultural versus theoretical. in phylogenetics, we study generalizations of HMMs (tree structured latent variable models) and typically use gradient based methods to fit the parameters, which is part of the reason i ask.

this referenced paper found the gradient based approach to have a faster rate of convergence than baum-welch near the optimum:

@uthsavc @naturemethods congrats! as someone who has seen this work presented many times, i attest that it is both quite elegant and a natural (and clever) way to model spatial transcriptomics data.

RT @uthsavc: GASTON, our method to learn “topographic maps” of gene expression, is out now @naturemethods! IMO the coolest part is a new m…

see also,

see, e.g. which compares the two approaches. why is the latter not more dominant?

@memecrashes fortunately, from the perspective of numerical integration, both are similarly easy to compute.

@miniapeur of course, if you prefer leetcode, do that instead. just try to enjoy the process ;)

@miniapeur problems are sorted by elo, so you can just choose problems in the elo range you are comfortable at.

@miniapeur slightly less directly since it has less data structure focus. that said, the leetcode problems become easy once you hit around ~1400-1500 cf elo

@Thomasdelvasto_ @pli_cachete computational/mathematical biology!

@pli_cachete @Thomasdelvasto_ i love my phd, definitely a skill issue.