For future reference: After trying different solutions (some are described in the above posts), the most performant solution was based on Michael Jones' pcg package ( https://github.com/MichaelTJones/pcg) with a tweaked Bound() function (to use an algorithm similar to Go's std lib). Using Go 1.12.0, this resulted in ~60% speedup (see benchmarks below). I have also tried a version of Ian Denhardt's solution but it was both slower and (as expected) inferior to the pcg package which on a sample of 10e6 runs produced a frequency distribution that deviated by less than 0.001 from the expected distribution.
Thanks again for everyone's help, Original implementation (based on Go std lib math/rand package) BenchmarkScalingFreqDistributionSampler/n-levels=2-12 50000000 32.1 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n-levels=4-12 50000000 35.4 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n-levels=8-12 50000000 32.0 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n-levels=16-12 50000000 32.0 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n-levels=32-12 50000000 32.1 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n-levels=64-12 50000000 36.9 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n-levels=128-12 50000000 31.6 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n-levels=256-12 50000000 31.0 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n-levels=512-12 50000000 36.1 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n-levels=1024-12 50000000 31.8 ns/op 0 B/op 0 allocs/op based on Michael Jones pcg package (https://github.com/MichaelTJones/pcg) with a tweaked Bound() function BenchmarkScalingFreqDistributionSampler/n_levels=2-12 100000000 11.3 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n_levels=4-12 100000000 14.1 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n_levels=8-12 100000000 11.3 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n_levels=16-12 100000000 11.6 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n_levels=32-12 100000000 11.9 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n_levels=64-12 100000000 15.5 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n_levels=128-12 100000000 11.5 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n_levels=256-12 100000000 11.2 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n_levels=512-12 100000000 15.2 ns/op 0 B/op 0 allocs/op BenchmarkScalingFreqDistributionSampler/n_levels=1024-12 100000000 11.9 ns/op 0 B/op 0 -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
