[ncl-talk] optimization problem

Mary Haley haley at ucar.edu
Tue Jul 26 10:55:09 MDT 2016


​Marston,

In general, I recommend trying to use array arithmetic over looping, even
if it means using extra memory to create 2D versions of 1D arrays.

If you have enough memory to create 2D arrays of your various 1D arrays,
then I recommend replacing all of the code after the last "copy_VarMeta"
call with the following (UNTESTED) code:

I'm not exactly sure I have all the dimensions of your various arrays
correct, so this code may need some correction. For example, I don't know
what the dimensions of "mhs" are, so I'm not sure that I have that part
right.  If mhs is already dimensioned ndarr x nchar, then the mhsx_2d lines
should look simply like this:

  mhs5_2d := round(mhs(4,:,:),0)


; Create 2D versions of these arrays for faster (?) calculations later

dims_2d = (/ndarr,nchar/)
mhs5_2d = conform_dims(dims_2d,round(ndtooned(mhs(4,:,:)),0),1)
mhs4_2d = conform_dims(dims_2d,round(ndtooned(mhs(3,:,:)),0),1)
mhs3_2d = conform_dims(dims_2d,round(ndtooned(mhs(2,:,:)),0),1)

tb5m4_2d = conform_dims(dims_2d,round(ndtooned(tb5m4),0),0)
tb5m3_2d = conform_dims(dims_2d,round(ndtooned(tb5m3),0),0)
tb4m3_2d = conform_dims(dims_2d,round(ndtooned(tb4m3),0),0)

darr_2d = conform_dims(dims_2d,darr,0)
char_2d = conform_dims(dims_2d,char,1)

; This creates a 2D logical array of Trues/Falses
mhs5_eq_char = mhs5_2d .eq. char_2d

; Create 2D arrays containing 1s and 0s
tmp5 = where(mhs5_eq_char .and. tb5m4_2d .eq. darr_2d, 1, 0)
tmp4 = where(mhs5_eq_char .and. tb5m3_2d .eq. darr_2d, 1, 0)
tmp3 = where(mhs5_eq_char .and. tb4m3_2d .eq. darr_2d, 1, 0)

hist5m4 = hist5m4 + tmp5
hist5m3 = hist5m3 + tmp4
hist4m3 = hist4m3 + tmp3​

delete([/mhs5_eq_char,
mhs5_2d,mhs4_2d,mhs3_2d,darr_2d,char_2d,tmp5,tmp4,tmp3/])

On Tue, Jul 26, 2016 at 6:05 AM, Marston Johnston <shejo284 at gmail.com>
wrote:

> Hi,
>
> I'm wondering if there is a way to speed up the do loops in the code below?
> I'm not sure it can be done, but I'm not an expert.
>
> ;***** 2D histogram ******
> ; x and y axis
> darr = ispan(-80,80,1)
> sarr = new(dimsizes(darr),string)
> sarr = " "
> char = ispan(150,300,1)
>
> ; histogram bins
> hist5m4 = new((/dimsizes(darr),dimsizes(char)/),float)
> hist5m4 at _FillValue = default_fillvalue("float")
> hist5m4!0 = "diff"
> hist5m4!1 = "Tb"
> hist5m4&diff = darr
> hist5m4&Tb = char
> ndarr = dimsizes(darr)
> nchar = dimsizes(char)
> hist5m4 = 0.0
> hist5m3 = hist5m4
> hist5m3 = 0.0
> hist4m3 = hist5m4
> hist4m3 = 0.0
>
> tb5m4 = mhs(4,:,:) - mhs(3,:,:)
> copy_VarMeta(mhs(3,:,:),tb5m4)
> tb5m3 = mhs(4,:,:) - mhs(2,:,:)
> copy_VarMeta(mhs(2,:,:),tb5m3)
> tb4m3 = mhs(3,:,:) - mhs(2,:,:)
> copy_VarMeta(mhs(2,:,:),tb5m3)
>
> mhs5 := round(ndtooned(mhs(4,:,:)),0)
> mhs4 := round(ndtooned(mhs(3,:,:)),0)
> mhs3 := round(ndtooned(mhs(2,:,:)),0)
>
> tb5m4 := round(ndtooned(tb5m4),0)
> tb5m3 := round(ndtooned(tb5m3),0)
> tb4m3 := round(ndtooned(tb4m3),0)
>
> ; print(darr(y)+" <= DARR:CH5AR => "+ch5ar(x))
> ; Fill the bins
> do x = 0, nchar - 1
>    do y = 0, ndarr - 1
>       tmp := ind(mhs5.eq.char(x) .and. tb5m4.eq.darr(y))
>       if(dimsizes(tmp).gt.1) then
>         hist5m4(y,x) = hist5m4(y,x) + (dimsizes(tmp) - 1)
>       end if
>       tmp := ind(mhs5.eq.char(x) .and. tb5m3.eq.darr(y))
>       if(dimsizes(tmp).gt.1) then
>         hist5m3(y,x) = hist5m3(y,x) + (dimsizes(tmp) - 1)
>       end if
>       tmp := ind(mhs4.eq.char(x) .and. tb4m3.eq.darr(y))
>       if(dimsizes(tmp).gt.1) then
>         hist4m3(y,x) = hist4m3(y,x) + (dimsizes(tmp) - 1)
>       end if
>   end do
> end do
>
> Appreciate your help with this,
> /M
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Marston S. Johnston, PhD
> Department of Earth Sciences
> University of Gothenburg, Sweden
> Email: marston.johnston at gu.se
> Phone: +46-31-7864901
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Only the fruitful thing is true!
>
> _______________________________________________
> ncl-talk mailing list
> ncl-talk at ucar.edu
> List instructions, subscriber options, unsubscribe:
> http://mailman.ucar.edu/mailman/listinfo/ncl-talk
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/ncl-talk/attachments/20160726/ea65c0bd/attachment.html 


More information about the ncl-talk mailing list