Timing

SpikeSorting.jl was built with online processing in mind, and is cable of sorting multiple channels simultaneously and taking advantage of multiple cores. How an algorithm scales across channels and cores can depend on the algorithm itself, your sampling rate and the machine it is running on. The simulations below demonstrate performance for various algorithms under some of these circumstances.

Simulation 1: Increasing channels and parallel processing



In [ ]:

    
addprocs()

using SpikeSorting, DistributedArrays, PyPlot

#Spike Sorting
s_s1=Array(Array{Sorting,1},64);
s_p1=Array(Any,64);

count1=1
for i=4:4:256
    #1 core
    s_s1[count1]=create_multi(DetectPower(),ClusterOSort(),AlignMax(),FeatureTime(),ReductionNone(),ThresholdMean(),i);

    #4 core
    s_p1[count1]=create_multi(DetectPower(),ClusterOSort(),AlignMax(),FeatureTime(),ReductionNone(),ThresholdMean(),i,2:5);
 
    count1+=1
end



In [ ]:

    
function timing1(mytimes::Array{Float64,2},s_s::Array{Array{Sorting,1},1},s_p)

    count=1
    for i=4:4:256
    
        #single core
        v=rand(1:1000,20000,i);
        (buf1,nums1)=output_buffer(i);
        cal!(s_s[count],v,buf1,nums1,true)
        cal!(s_s[count],v,buf1,nums1);
    
        t1 = @elapsed onlinesort!(s_s[count],v,buf1,nums1);
    
        for j=1:20
            t1 = @elapsed onlinesort!(s_s[count],v,buf1,nums1);
            if t1<mytimes[count,1]
                mytimes[count,1]=t1
            end
        end
    
        v2=convert(SharedArray{Int64,2},v);
        (buf2,nums2)=output_buffer(i,true);
        cal!(s_p[count],v2,buf2,nums2,true);
        cal!(s_p[count],v2,buf2,nums2);
    
        t2 = @elapsed onlinesort!(s_p[count],v2,buf2,nums2);
    
        for j=1:20
            t2 = @elapsed onlinesort!(s_p[count],v2,buf2,nums2);
            if t2<mytimes[count,2]
                mytimes[count,2]=t2
            end
        end
    
        count+=1

    end

    nothing
    
end



In [ ]:

    
times=ones(Float64,64,2)

timing1(times,s_s1,s_p1);



In [ ]:

    
(fig,myax)=subplots(1,1)
plot(4:4:256,times[:,1].*1000,4:4:256,times[:,2].*1000)
myax[:set_xlabel]("Number of Channels")
myax[:set_ylabel]("Time (ms)")
myax[:legend](["1 Core", "4 Core"], loc=2)



In [ ]: