6. Find Best Stats for Model Choice

Use ABCtoolbox task findStatsModelChoice on 100,000 simulations, 1000 retained simulations, 100 cross validation.

ABCtoolbox test_ABC_findStatsModelChoice3.txt

Where test_ABC_findStatsModelChoice3.txt is:

task findStatsModelChoice
simName pruneCorStats_input_ABCtoolbox_M1_HPC_OSG_203.txt;pruneCorStats_input_ABCtoolbox_M2_HPC_OSG.txt;pruneCorStats_input_ABCtoolbox_M3_HPC_OSG.txt
obsName pruneCorStats_real_output_M23.summary
params 2-6,8-14,16-22;2-6,8-15,17-24;2-6,8-16,18-26
numRetained 1000
maxReadSims 100000
modelChoiceValidation 100
diracPeakWidth 0.01
posteriorDensityPoints 100
standardizeStats 0
writeRetained 1
outputPrefix ABC_searchStatsForModelChoice_pruneCorStats_100000_1000_100_
pruneCorrelatedStats 1
logFile ABC_searchStatsForModelChoice_pruneCorStats_100000_1000_100.log
verbose

If ABCtoolbox is compled with openMP, findStatsModelChoice can be run on the number of cores equal to the number of models. The number of statistics is the main component in run time. This is why we prune for highly correlated summary statistics before we find the best statistics for model choice. With 54 summary statistics, 100,000 simulations, 1000 retained simulations, and 100 cross validations it took 1148.7 min to complete.