[ncl-talk] regCoef_n without y-axis intercept

Dennis Shea shea at ucar.edu
Thu Oct 27 11:09:37 MDT 2022 

This thread is a bit hard to follow for me.

[1] re: "Now I'm thinking how to make the function to force the regression
to pass from origin. Any idea on how to do that?"
     I do not know how to accommodate the additional constraint of making
the regression line'go through the origin.
     Likely, googling something like "regression line through the origin"
will yield some information.

[2] Make sure your modified fortran subroutine works. Initially, code and
test this only for one-dimension. We can do multiple dimensions later.
    EG with the 'gfortran' compiler.
     I took NCL's underlying fortran subroutine. I changed the source name
to RCGRAF/rcgraf to avoid any possible naming conflicts. I wrote a simple
fortran driver. Any  fortran compiler will work

     %> gfortran tst_regcoef.f regcoef_graffino.f        <=== compile
fortran codes
      %> ./a.out
<=== execute the compiled codes

       rcoef=  0.974561453
        tval =   38.7428589
        yint =   15.3522949

[3] Now, modify the "regcoef_graffino.f  " to create a regression line
where there is no y-intercept

[4] Repeat step [2] with your modified code.

[5] Create a .so file via

     %> WRAPIT regcoef_graffino.f

      will create regcoef_graffino.so

[6] Create a simple 1D test of invoking the .so file and print results

[7] At this point, we can make a 2nd subroutine that invokes the .so file
multiple times.











If The (hidden) C interface to the f77 subroutine just loops over the
appropriate dimensions and returns the various results.

---
If you have 1D arrays [ X[N], Y[N] ]


On Wed, Oct 26, 2022 at 9:10 AM Dave Allured - NOAA Affiliate via ncl-talk <
ncl-talk at mailman.ucar.edu> wrote:

> Giorgio, WRAPIT is capable of handling arrays correctly.  I suggest you
> try some of the working fortran examples in the WRAPIT documentation, so
> that you can get familiar with WRAPIT and array arguments.  Then, get the
> *original* regcoef_dp.f working with WRAPIT and arrays, but change the
> function name to prevent conflict.  It might be helpful to write external
> standalone fortran main programs that call the fortran routines you are
> testing, to ensure those routines are working independently from NCL.
>
> Only after that, try interfacing your modified routine with WRAPIT.
> Please completely avoid regcoefW.c.  It is full of internal protocol that
> is not related to WRAPIT and will continue to confuse you.
>
> I do not have time to analyze or debug your code and procedures.  If
> needed, perhaps someone else is available.
>
>
> On Wed, Oct 26, 2022 at 3:09 AM Giorgio Graffino <g.graffino at tim.it>
> wrote:
>
>> Hi Dave,
>>
>> Thanks for confirming that. I'm changing the fortran subroutine Dennis
>> sent last week into a function, but it keeps giving me an error wherever I
>> try to use arrays as arguments. It only works when X and Y are scalar
>> numbers. I really don't understand how a linear regression fit can be found
>> with scalar numbers as arguments. I'm looking at the C interface to see how
>> to subroutine is called, but there are too many things happening in
>> regcoefW.c and it's difficult to understand the workflow. Can anyone please
>> help me understand how to make the fortran subroutine work for 1d and 2d
>> arrays?
>>
>> Cheers,
>>
>> Giorgio
>>
>>
>> ------ Messaggio Originale ------
>> Da: dave.allured at noaa.gov
>> A: g.graffino at tim.it
>> Cc: ncl-talk at ucar.edu
>> Inviato: lunedì 24 ottobre 2022 22:10
>> Oggetto: Re: [ncl-talk] regCoef_n without y-axis intercept
>>
>> Giorgio, you are right, regcoefW.c is an NCL internal interface routine
>> for regCoef_n. But it uses a different protocol than a WRAPIT interfaced
>> function. Just ignore regcoefW.c, and follow the WRAPIT fortran
>> instructions carefully, for your new function. Change the function name,
>> and do not attempt to replace the built-in regCoef_n function. You can
>> change the arguments as you like, but follow the documented WRAPIT rules.
>>
>> There are two different methods for calling external C functions: fortran
>> wrapper, and shared object. Please study the WRAPIT documentation carefully.
>>
>>
>> On Mon, Oct 24, 2022 at 1:35 PM Giorgio Graffino via ncl-talk <
>> ncl-talk at mailman.ucar.edu> wrote:
>>
>>> Hello,
>>>
>>> I'm doing some modifications to the regcoef_dp.f file to consider a
>>> zero y-intercept, but now I don't know how to try it in a script. The
>>> number of arguments in the fortran subroutine is different than in
>>> regCoef_n, so the subroutine is probably going through some more steps,
>>> possibly involving this script (
>>> https://github.com/NCAR/ncl/blob/develop/ni/src/lib/nfp/regcoefW.c). Am
>>> I right?
>>>
>>> How can I incorporate both fortran and C scripts in my NCL script? I
>>> know I can use WRAPIT to read external fortran code, but what about C code?
>>>
>>> I need help please.
>>>
>>> Giorgio
>>>
>>>
>>> ------ Messaggio Originale ------
>>> Da: ncl-talk at mailman.ucar.edu
>>> A: shea at ucar.edu
>>> Cc: ncl-talk at ucar.edu
>>> Inviato: venerdì 21 ottobre 2022 13:14
>>> Oggetto: Re: [ncl-talk] regCoef_n without y-axis intercept
>>>
>>> Hi Dennis, Dave
>>>
>>> Thanks for sending the relevant function and for explaining how to
>>> access the source code. Now I'm thinking how to make the function to force
>>> the regression to pass from origin. Any idea on how to do that?
>>>
>>> Cheers,
>>>
>>> Giorgio
>>>
>>>
>>> ------ Messaggio Originale ------
>>> Da: shea at ucar.edu
>>> A: g.graffino at tim.it
>>> Cc: dave.allured at noaa.gov; ncl-talk at ucar.edu
>>> Inviato: giovedì 20 ottobre 2022 19:17
>>> Oggetto: Re: [ncl-talk] regCoef_n without y-axis intercept
>>>
>>> The f77 code is attached.
>>>
>>>
>>> On Thu, Oct 20, 2022 at 9:53 AM Giorgio Graffino via ncl-talk <
>>> ncl-talk at mailman.ucar.edu> wrote:
>>>
>>>> Hi Dave,
>>>>
>>>> Thanks for the advice. I looked at the examples in
>>>> https://www.ncl.ucar.edu/Applications/regress.shtml, especially
>>>> examples 2 and 4, but I couldn't find any supplementary information about
>>>> regCoef. I also can't find anything under
>>>> $NCARG_ROOT/lib/ncarg/nclscripts/csm/gsn_code.ncl and
>>>> $NCARG_ROOT/lib/ncarg/nclscripts/csm/contributed.ncl.
>>>>
>>>> Is NCL source code located anywhere in $NCARG_ROOT? Can anyone please
>>>> point me where it is?
>>>>
>>>> Cheers,
>>>>
>>>> Giorgio
>>>>
>>>>
>>>> ------ Messaggio Originale ------
>>>> Da: dave.allured at noaa.gov
>>>> A: g.graffino at tim.it
>>>> Cc: ncl-talk at ucar.edu
>>>> Inviato: giovedì 13 ottobre 2022 18:47
>>>> Oggetto: Re: [ncl-talk] regCoef_n without y-axis intercept
>>>>
>>>> Giorgio, for your second question, please see this very brief
>>>> description that I recently wrote, for how to search for functions in the
>>>> NCL source code.
>>>> https://mailman.ucar.edu/pipermail/ncl-talk/2022-August/018954.html
>>>>
>>>>
>>>> On Thu, Oct 13, 2022 at 10:38 AM Giorgio Graffino via ncl-talk <
>>>> ncl-talk at mailman.ucar.edu> wrote:
>>>>
>>>>> Hi NCL community,
>>>>>
>>>>> I want regCoef_n to find the regression coefficient between two arrays
>>>>> without y-axis intercept. Do you know how I can change the function to do
>>>>> that?
>>>>>
>>>>> Can you also please tell me where I can find the function? I did a
>>>>> grep recursive search in my $NCARG_ROOT directory but I couldn't find
>>>>> it.
>>>>>
>>>>> Cheers,
>>>>>
>>>>> Giorgio
>>>>>
>>>> _______________________________________________
> ncl-talk mailing list
> ncl-talk at mailman.ucar.edu
> List instructions, subscriber options, unsubscribe:
> https://mailman.ucar.edu/mailman/listinfo/ncl-talk
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.ucar.edu/pipermail/ncl-talk/attachments/20221027/903cf040/attachment.htm>

More information about the ncl-talk mailing list