[ncl-talk] index lists in assignment

Fri Nov 30 16:30:49 MST 2018

Hi Axel,

*NCL 6.5.0 <http://www.ncl.ucar.edu/current_release.shtml>* has a task
parallelism feature created by Rick.

Task Parallelism Examples are at:

*http://www.ncl.ucar.edu/Applications/task_parallelism.shtml*
<http://www.ncl.ucar.edu/Applications/task_parallelism.shtml>

I wonder if this might be of use.

Cheers
D

On Fri, Nov 30, 2018 at 2:09 PM Seifert Axel <Axel.Seifert at dwd.de> wrote:

>
> Hi Dennis and Rick,
>
>
> Thanks for your comments. Yes, the do-loop is kind of slow and would then
> be the most expensive part of that script. The length of an observation
> vector is of order 100000 for one hour of satellite data and I have
> several months of data, which have to be processed for multiple model
> simulations.
>
>
> I can probably restructure my data to avoid double indices, or multiple
> overpasses, in one observation vector. Alternatively, I could simply ignore
> that slight loss of data. At least, it would always use the newer data and
> overwrite the older data. This is probably not a bad choice as well.
>
>
> Axel
>
>
>
>
> ------------------------------
> *Von:* Dennis Shea <shea at ucar.edu>
> *Gesendet:* Freitag, 30. November 2018 20:21
> *An:* Rick Brownrigg
> *Cc:* Axel.Seifert at dwd.de; Ncl-talk
> *Betreff:* Re: [ncl-talk] index lists in assignment
>
> My recollection is that the way fortran and NCL handled 'indirect
> subscripting' (my words) yielded different results.
> Hence, the 'expected results' are colored by past expeerience in fortran.
>
> Unfortunately, I can not find my NCL/fortran   example. Sorry.
>
> D
>
>
> On Fri, Nov 30, 2018 at 11:27 AM Rick Brownrigg <brownrig at ucar.edu> wrote:
>
>> Hi Axel,
>>
>> Apologies in delay in getting back to you on this.  As far as I can tell,
>> the behavior ox "x(idx) = x(idx) + z" is the intended behavior (perhaps not
>> the same as "desired" behavior in your case), in the sense that the
>> right-hand side is completely evaluated before the (re)assignment to the
>> variable "x": is performed.
>>
>> The do-loop approach does work -- it it just unbearably slow?  Without
>> knowing more about your data, I don't have a suggestion for how to avoid
>> looping.
>>
>> Rick
>>
>>
>> On Thu, Nov 29, 2018 at 10:37 AM Axel Seifert <Axel.Seifert at dwd.de>
>> wrote:
>>
>>>
>>> Maybe I should add that the output of my little script is
>>>
>>>
>>> (0)      0 0
>>> (1)      3 4
>>> (2)      2 2
>>>
>>>
>>> This is with NCAR Command Language Version 6.3.0
>>>
>>>
>>> The variables x and y are
>>>
>>>
>>> Variable: x
>>> Type: float
>>> Total Size: 12 bytes
>>>             3 values
>>> Number of Dimensions: 1
>>> Dimensions and sizes:   [3]
>>> Coordinates:
>>> Number Of Attributes: 1
>>>   _FillValue :  9.96921e+36
>>> (0)      0
>>> (1)      3
>>> (2)      2
>>>
>>> Variable: y
>>> Type: float
>>> Total Size: 12 bytes
>>>             3 values
>>> Number of Dimensions: 1
>>> Dimensions and sizes:   [3]
>>> Coordinates:
>>> Number Of Attributes: 1
>>>   _FillValue :  9.96921e+36
>>> (0)      0
>>> (1)      4
>>> (2)      2
>>>
>>>
>>>
>>> Am 24.11.2018 um 15:16 schrieb Seifert Axel:
>>>
>>>
>>> Hi,
>>>
>>>
>>> I have a somewhat unexpected behavior when using index lists. Here a
>>> simplified example:
>>>
>>>   x = new( 3, float)
>>>   x = 0.0
>>>   y = x
>>>
>>>   z = (/1.,2.,3./)
>>>   idx = (/1,2,1/)
>>>
>>>   x(idx) = x(idx) + z
>>>
>>>   do i=0,dimsizes(idx)-1
>>>     y(idx(i)) = y(idx(i)) + z(i)
>>>   end do
>>>
>>>   print(" "+x+" "+y)
>>>
>>> I would have expected that x and y end up being the same, but they are
>>> not. The double assignment to index 1 does not work for the vector
>>> assignment. Is this the intended behavior for arrays in ncl?
>>>
>>> How can I get the 2nd behavior and still avoid the do-loop?
>>>
>>> In my case I am assigning measured values from satellite overpasses to a
>>> global grid and it can of course happen that I have multiple overpasses for
>>> a grid point in an observation vector. Hence, in addition to the sum of the
>>> values I would count the number of overpasses
>>>
>>>   cnt = (/0,0,0/)
>>>   cnt(idx) = cnt(idx) + 1
>>>   print(" "+cnt)
>>>
>>> and then xavg = x/cnt should give me the proper average for, for
>>> example, one hour of data. What I need in the end is the correct average
>>> for xavg and the number of measurements used at each grid point.
>>>
>>> Thanks in advance for any helpful comments and suggestions.
>>>
>>> Axel
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>> _______________________________________________
>>> ncl-talk mailing listncl-talk at ucar.edu
>>> List instructions, subscriber options, unsubscribe:http://mailman.ucar.edu/mailman/listinfo/ncl-talk <https://ssi.dwd.de/mailman/listinfo/,DanaInfo=.ambkoqftG2lk2Lrr9+ncl-talk>
>>>
>>>
>>> _______________________________________________
>>> ncl-talk mailing list
>>> ncl-talk at ucar.edu
>>> List instructions, subscriber options, unsubscribe:
>>> http://mailman.ucar.edu/mailman/listinfo/ncl-talk
>>> <https://ssi.dwd.de/mailman/listinfo/,DanaInfo=.ambkoqftG2lk2Lrr9+ncl-talk>
>>>
>> _______________________________________________
>> ncl-talk mailing list
>> ncl-talk at ucar.edu
>> List instructions, subscriber options, unsubscribe:
>> http://mailman.ucar.edu/mailman/listinfo/ncl-talk
>> <https://ssi.dwd.de/mailman/listinfo/,DanaInfo=.ambkoqftG2lk2Lrr9+ncl-talk>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.ucar.edu/pipermail/ncl-talk/attachments/20181130/0d7e5855/attachment.html>