[Ifeffit] Multiple data set fit limit

Fri Mar 24 23:37:28 CDT 2017

Bruce,
Thank you for your reply and suggestions.

1. In regard to the suggestion 1, my real problem (refinement) has 3 data sets and the model is rather complex with 55 paths. For some reasons,  Artemis (when project generated under Windows and is opened under Linux) remembers previous 3 data sets (replaced with the new ones) and generated 330 paths (6 x 55paths) instead of 165. I thought that there is a  limit in the amount of paths? To test that I have generated a large number of copies of the data with simpler model (FeS2). When the number of path was well over 300 (more than 16 data sets), I encountered the problem reported.

2. I have tried to run Larch, that started Larch server and then run Artemis. I don't see any indication that the Artemis uses Larch. How to make Artemis to use Larch instead of Ifeffit?

Thanks again,

Victor

-----Original Message-----
From: Ifeffit [mailto:ifeffit-bounces at millenia.cars.aps.anl.gov] On Behalf Of Bruce Ravel
Sent: Saturday, 25 March 2017 4:17 AM
To: XAFS Analysis using Ifeffit <ifeffit at millenia.cars.aps.anl.gov>
Subject: Re: [Ifeffit] Multiple data set fit limit

Victor,

I now believe that you are seeing the problem you claim to be seeing, but it's going to take a lot of work for me to understand how you got there.

When I import the Artemis project files you sent, I see that you have 24 copies of the FeS2 data in the project and you seem to be including 3 of them in the fit.

I suspect the central problem here is that Demeter is making an ifeffit group for each copy of the data, even the ones you are excluding from the fit.  This exceeds the normal compiled-in limit of 16.  From a user interface perspective, I suspect that the sin is that Demeter blithely does this without warning you about Ifeffit's limit of 16 data sets.

It is going to take me a long time to really track down all the things you have done to push at the boundaries of what Artemis + Ifeffit can do.  I have not done much testing beyond a handful of data sets (neighborhood of 6).  It's possible I have done no testing in double digits and I have clearly not done much testing up against Ifeffit's compiled-in limitation on number of data sets.

Some suggestions:

1. Don't exceed 16 data sets, even if you are planning on clicking off the "include" button for some of them.  In earlier email, when I said I do not see what you are observing, what I really meant was that if I do a fit with a handful of data sets, all of which are included in the fit, I see no such problem.

2. Try using Larch.  Larch, not being a Fortran program, does not have Ifeffit's compiled in limits.  That is not to say that Artemis+Larch will certainly work correctly with a large number of data sets (some of which are excluded from the fit) -- there most certainly could still be bugs in this untested regime.

In any case, there are things that could be done to make it easier to identify the exact problem you are uncovering.  I can see from examining your project files that you added a large number of copies of the data then hit the Fit button a couple dozen times before finally saving and sending me the project file.  From my perspective as developer and bug-squisher, it is always better to see an example that is as minimal as possible.  The more data you add after seeing a problem, the more times you hit the fit button after seeing a problem, the harder it is for me to track down the actual proximate cause of problems.

I acknowledge that this was likely not the answer you were hoping for. 
I guess I am saying, take small steps as you build up a big, complex fitting model.  It is true that a program written by a competent programmer should be able to take whatever you throw at it.  Alas, Artemis was written by a dilettante.

B

On 03/18/2017 10:59 PM, Victor Streltsov wrote:
> Hi Bruce,
> Sorry for delayed reply. I didn't want to take up too much of your time and I have attempted to run more tests to reproduce the problem with other data and models.
>
> First, I thought that you were able to reproduce the problem (maybe partially) since you wrote:
>> On 03/08/2017 10:53 AM, Bruce Ravel wrote:
>> Well ..... a multiple data set fit using larch runs to completion and 
>> reports sensible values for parameters, but does not manage the data 
>> sets correctly.  One obvious sign that something has gone wrong is 
>> the after-fit plot attached.  Yikes!
>
> Similar behaviour I observed in my refinement using Ubuntu version of 
> Demeter/Artemis/Ifeffit. Windows version works fine. (Switching to 
> Linux version is related to my previous bug report about user input of 
> epsilon which is still not available under Windows (until next 
> release), but fixed in the latest Linux version(s))
>
> The task I was running included 3 data sets and a model with 55 paths. However, the "Monitor/Debug options/Show mode status" in Demeter/Artemis shows content with 330 paths instead of expected 3x55= 165, if I understand correctly this option. It seems that it counts all added and previous replaced/removed data sets (extra 3 datasets which were not included and even removed). It doesn't effect Windows based run, however, creates the problem I mentioned before using the same project file under Linux. Is number of paths a compiled-in limit (less than ~300 or so) in Linux/Ifeffit?
>
> To avoid any possible problems caused by my model and data I used available for Artemis example FeS2 data for test runs.  I have simply added many the same FeS2 data sets in the Artemis but included only 3 data sets to actual fit. Extra unused data sets did not affect at all the fit of 3 data sets under Windows, however, the problem reproduced using Linux version. I have attached (zip) project files, log files and contents file (mode status) for both runs under Windows and Linux.
>
> I hope it is more helpful.
>
> Victor
>
> -----Original Message-----
> From: Ifeffit [mailto:ifeffit-bounces at millenia.cars.aps.anl.gov] On 
> Behalf Of Bruce Ravel
> Sent: Saturday, 11 March 2017 1:17 AM
> To: XAFS Analysis using Ifeffit <ifeffit at millenia.cars.aps.anl.gov>
> Subject: Re: [Ifeffit] Multiple data set fit limit
>
>
> On 03/09/2017 05:15 AM, Victor Streltsov wrote:
>> 3 data set refined to completion with sensible refined parameter, 
>> however,
>>
>> R-factor for 3^rd data set is 1.0000 (overall R for 3 data set is 
>> huge
>> too) and after-fit plot for 3^rd data set is missing, only 
>> experimental one shown. I am using latest versions of programs.
>>
>
> Again, I don't see this behavior on my own computer.
>
> It is always better to /show/ something to me than to /explain/ something to me.  If you are unwilling or unable to make an actionable bug report, then there isn't anything that I am going to be able to do for you.
>
> It is not my intent to be difficult.  You are telling me about things that I cannot reproduce on my own computer.  Despite my explicit request for a more useful bug report, you are not showing me how to see what you are seeing.
>
> If you cannot do something to make this conversation more productive, then I am just going to stop responding.
>
> http://bruceravel.github.io/demeter/documents/SinglePage/bugs.html
>
> B
>
>
>
>
> --
>   Bruce Ravel  ------------------------------------ bravel at bnl.gov
>
>   National Institute of Standards and Technology
>   Synchrotron Science Group at NSLS-II
>   Building 743, Room 114
>   Upton NY, 11973
>
>   Homepage:    http://bruceravel.github.io/home/
>   Software:    https://github.com/bruceravel
>   Demeter:     http://bruceravel.github.io/demeter/
> _______________________________________________
> Ifeffit mailing list
> Ifeffit at millenia.cars.aps.anl.gov
> http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit.
> aps.anl.gov
> Unsubscribe: 
> http://millenia.cars.aps.anl.gov/mailman/options/ifeffit.
> aps.anl.gov
>
>
>
> _______________________________________________
> Ifeffit mailing list
> Ifeffit at millenia.cars.aps.anl.gov
> http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit.
> aps.anl.gov
> Unsubscribe: 
> http://millenia.cars.aps.anl.gov/mailman/options/ifeffit.
> aps.anl.gov
>

--
  Bruce Ravel  ------------------------------------ bravel at bnl.gov

  National Institute of Standards and Technology
  Synchrotron Science Group at NSLS-II
  Building 743, Room 114
  Upton NY, 11973

  Homepage:    http://bruceravel.github.io/home/
  Software:    https://github.com/bruceravel
  Demeter:     http://bruceravel.github.io/demeter/
_______________________________________________
Ifeffit mailing list
Ifeffit at millenia.cars.aps.anl.gov
http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit
Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffit