No, I haven't used the profiler. However, I already know that most time is
spend in the two parts that are @time-ed. If you put a @time in front of
main() you get something like (I'm using Julia 4.3-pre6 on Debian):
julia> include("testnpreg.jl")
1.228517 seconds (34.96 k allocations: 764.159 MB, 6.98% gc time)
1.837348 seconds (234.88 k allocations: 1.432 GB, 8.50% gc time)
3.281414 seconds (317.79 k allocations: 2.331 GB, 7.41% gc time)
julia>
The first is the time to compute weights, the second the time for npreg,
and the third is total for main(). So you can see that the sum of the two
parts I'm timing is the majority of total time. I am not concerned with
reducing the time of computing the weights, at the moment. I would like to
optimize the time of the call to npreg (the second of the above) for now.
On Thursday, December 17, 2015 at 1:39:47 PM UTC+1, Patrick Kofod Mogensen
wrote:
>
> Did you try to use the profiler? Where is the time spent?
