simultaneous fitting?

[Dan Parshall]

I'm using Fityk 1.0.1 for linux. 

I'm working with data which have scans taken under different conditions.  There should be the same number of peaks in each scan, and their centers and widths should be fixed across all scans, but their intensities should change.

There seems to be some bug, in that sometimes when trying to fit multiple datasets, fityk only fits one dataset, or one at a time, or something.  It's fairly reproducible, in that I can generate a script which reliably produces this effect.  I haven't been able to figure out what triggers this, though.

I'm pretty sure that I have my variables linked correctly, because fitting a single dataset will produce changes to the models in other scans (i.e., the centers and widths will be different).  But selecting all the scans/ datasets and trying to fit simultaneously results in only the last scan being fit.

Help?
--
Dan Parshall
Shorty George Productions
Boulder, CO

[Christian Weikusat]

Sharing your script might help us to understand your problem!

Cheers,
Christian

[Dan Parshall]

Attached, along with the datasets.  Thanks!

--
http://groups.google.com/group/fityk-users

[Marcin Wojdyr]

There is no fit command in the script.
Do you use: "fit @*" ?

Marcin

[Dan Parshall]

yes.  whether I use the GUI, or:
@* : fit
@0 @1 : fit

The result is the same :/

Marcin

--
http://groups.google.com/group/fityk-users

[Marcin Wojdyr]

Try:
fit @*
The difference is that "fit @*" fits all datasets simultaneously,
while "@*: fit" fits all datasets one by one, separately.
The former takes more time but it's what you need.
Marcin

> --
> http://groups.google.com/group/fityk-users

[Dan Parshall]

well, that worked.  Is this behavior different from earlier versions of fityk?  I recall being able to select multiple datasets and fit them simultaneously (actually, this was one of the key reasons I started working with fityk) ...

I strongly suggest you make this the default behavior in the future.  If the user wants to fit things one at a time, they can go through the datasets one-by-one.  But selecting multiple datasets should mean that they'll all be fit together.

[Marcin Wojdyr]

Ok, I've just changed it. I've added a simple check if the fitted
models in different datasets use the same variables.
If they use, they are fitted together. Otherwise, one by one, which
should be faster.

Marcin

[Dan Parshall]

Excellent, thank you.  I suspect that will be both more intuitive, and more flexible, for future users.

On a related note, is it possible to calculate the errors in this situation?  In my case, I'm fitting several peaks across multiple datasets.  When I try
@* : info peaks_err

the errors reported are, I believe, calculated for each individual dataset.  Because I'm fitting across many datasets, I feel like the error should be much lower than that reported.

Marcin

--
http://groups.google.com/group/fityk-users

[Marcin Wojdyr]

You're right. That's one of two problems with "info peaks_err", the
other is output format.
OTOH if all datasets were included when calculating covariance matrix
and errors, it would be confusing for people who fit independent
datasets, because one dataset would affect errors in others
You may use Fit > Info and "info errors" to see uncertainties
calculated for all selected datasets together.
Marcin

> --
> http://groups.google.com/group/fityk-users

[Dan Parshall]

Hi Marcin-

I'm following up on this earlier question.  If I use the command:
@*: info errors >> fileout_errordata

the result seems to be the error calculated for each separate scan. 

Is there a way to get the composite error from the command line?  That is, can I get the same result that comes by selecting all the datasets and using the menu option Fit > Info ?  And most of all, can I output it to a file?  While "info errors @1 >> fileout" works for an individual file, "info errors @* >> fileout" yields an error.

Thanks!

[Marcin Wojdyr]

Dan,
sorry for late reply.

> While "info errors @1 >> fileout" works for an individual file, "info errors
> @* >> fileout" yields an error.

This problem should be fixed in ver. 1.2.0
https://github.com/wojdyr/fityk/commit/f12c677b8d9ac5e90ede9bcf11f6c9f7cfa23ede

Marcin