Milan,
Thanks for this! That is quite a difference, I might be making some
changes to my code!
Chris
On Thursday, January 28, 2016 at 5:36:07 AM UTC-5, Milan Bouchet-Valat
wrote:
>
> Le mercredi 27 janvier 2016 à 22:37 -0800, Christopher Alexander a
> écrit :
> > Try something like this:
> >
> > adjCloseMuVec = Vector{Float32}(nFullYears)
> > or
> > adjCloseMuVec = zeros(Float32, nFullYears)
> >
> > This will init a vector of the proper size. I think zeros is
> > slightly faster, but someone can correct me if that is not the case.
> Actually that's the other way around. zeros() is slower because it
> needs to fill the array with zeros. You can check this with some very
> simple benchmarking:
>
> julia> f() = for i in 1:100000 adjCloseMuVec = Vector{Float32}(10000) end
> f (generic function with 1 method)
>
> julia> g() = for i in 1:100000 adjCloseMuVec = zeros(Float32, 10000) end
> g (generic function with 1 method)
>
> julia> @time f()
> 0.164794 seconds (202.00 k allocations: 3.731 GB, 52.40% gc time)
>
> julia> @time f()
> 0.135390 seconds (200.00 k allocations: 3.731 GB, 52.57% gc time)
>
> julia> @time g()
> 1.184917 seconds (201.75 k allocations: 3.731 GB, 10.57% gc time)
>
> julia> @time g()
> 1.175267 seconds (200.00 k allocations: 3.731 GB, 10.38% gc time)
>
>
> This overhead could be reduced/eliminated by using calloc, and could
> even become the default, see
> https://github.com/JuliaLang/julia/issues/9147
>
> Given how high is the overhead, it looks like that change would be a
> good idea.
>
>
> Regards
>
>
> > Chris
> >
> > On Thursday, January 28, 2016 at 1:26:09 AM UTC-5, Michael Landis
> > wrote:
> > > I am trying to compute some historical means and volatilities with
> > > Julia, but I've been having some difficulty getting the
> > > declarations and/or assignments right...
> > >
> > > When I try a declaration, like either of these:
> > > adjCloseMuVec::Vector{Float32}; # or...
> > > adjCloseMuVec::Vector{Float32,nFullYears};
> > > I get: UndefVarError: adjCloseMuVec not defined
> > >
> > > When I try:
> > > adjCloseMuVec = Vector{Float32}; # and later try:
> > > reshape(adjCloseMuVec,nFullYears);
> > > I get: MethodError: `reshape` has no method matching
> > > reshape(::Type{Array{Float32,1}}, ::Int32)
> > >
> > > When I try:
> > > adjCloseMuVec = Vector{Float32,nFullYears};
> > > I get: too many parameters for type typealias
> > >
> > > So, the only acceptable path would seem to be:
> > > adjCloseMuVec = Vector{Float32};
> > > but when I later try:
> > > adjCloseMuVec[curYr] = Float32(adjCloseMu); # or...
> > > setindex!(adjCloseMuVec, Float32(adjCloseMu), curYr);
> > > I get: MethodError: `setIndex!` has no method matching
> > > setindex!(::Type{Array{Float32,1}}, ::Float32, ::Int64)
> > >
> > > So, I'm mystified as to the correct syntax. Any suggestions on an
> > > acceptable way to declare a Vector of Int32 and load values into
> > > it?
> > >
>
