Abc For Gpas Part2

ABC for GPAS resource optimization (part 2)

[2019-11-19]

But dang! The resulting parameter space defined in by matrix with 4 columns (1 for each parameter) and 796,902 rows (seq(from=0, to=1, by=0.01) is too big, my sampling function and R are too slow to iterate across these times 100 replications! So now I’ll try to do this in Julia with parallel computation.

I’ve implemented Julia’s Distributed standard library instead of GNU parallel, and it’s not as painful as when I first tried it a few months ago. The trick is to use the macro @everywhere to define variables and functions on all threads and that the number of parallel threads are defined via julia -p${nCores} or within the script via Distributed.addprocs(nCores); while defining the bash variable export JULIA_NUM_THREADS=${nCores} does not work for some reason I’m too lazy to find out.

The repo I’m pushing into: genomic_prediction.git with a production-ready repo for Pool-GWAS.

SIDE NOTES:

• Excluding missing and NaN data types from calculations in Julia1.0.1 are frustratingly inelegant!
• dropmissing(df), skipmissing(array), ismissing(array), and .!isnan.(array) are just too many different functions for something as common as missing values in data analysis.