I am using RCall.jl to run some estimations of hidden markov models using the R package
seqHMM. I have 64GB of RAM and task manager never shows me using above 20GB total, yet I am getting out of memory errors when estimating the model. The data itself is small, about 4,000 sequences with the longest one being 7,000 observations. I’d guess this should take no more than a 100Mb to store that. I am using 16 threads for the estimation. This is all within a Pluto notebook, if that makes a difference.
Are there certain limits that RCall sets that I just need to adjust? I just don’t understand how I can possibly be running out of memory (especially when it doesn’t even appear like I am!).
PS. Bonus points if someone has secretly written an HMM package in Julia (I couldn’t find any).