Re: [Psignifit-users] Error in Matlab interface

[Ingo F. <ing...@tu...>]

Dear Johahn,

you are right, there is not sensitivity analysis yet in the matlab code.
I added that as a todo.

With respect to the deviance, you should be able to also get deviances
for all the bootstrap samples. If you use the GoodnessOfFit plot, you
will see the deviance of the ML-estimate along with a histogram of the
deviances from the bootstrap samples. That should help you interpret the
deviance value you get.

Best,

Ingo

On Fri, Jun 10, 2011 at 04:57:31PM +1000, Johahn Leung wrote:
>    Hi Ingo,
>    Thanks, I just downloaded the newest version (07_06) and the code runs.
>    However, I note that the Matlab interface does not have a Sensitivity
>    analysis, or is that incorporated in the getCI function?
>    I am getting really small CI and have no way of telling if it's correct. 
>    Using GoodnessofFit, i can get a value of deviance, but again, need a way
>    to interpret it.
>    cheers
>    /j
>    ___________________________
>    Johahn Leung
>    Auditory Neuroscience Laboratory
>    Department of Physiology
>    University of Sydney
>    T +612 93517615
>    F +612 93518259
> 
>    On Tuesday, 7 June 2011 at 8:01 PM, Ingo FrA 1/4nd wrote:
> 
>      Dear Johahn,
> 
>      the fix has been uploaded. The other bug you report below should have
>      been fixed
>      some time ago. See commit:
> 
>      833a8e9095aebb4ed907729f9fa0d3bf245a4653
> 
>      and in particular:
>      f99cd7e169c87f2821a336714e69162c78742969
> 
>      Which version are you using? If you are experienceing these problems
>      with a recent snapshot of psignifit3 could you please write me a short
>      note? What you report should not happen. We have implemented another
>      strategy for sampling from the posterior: The ASIRInference object in
>      python tries to find a reasonable parametric approximation to the
>      posterior marginals and then uses sampling importance resampling. This
>      way, there are no issues about chain convergence anymore. However, ASIR
>      inference is not yet available from matlab. If you are interested in
>      trying that out, please drop me a note.
> 
>      Thanks a lot,
> 
>      Ingo
> 
>      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>      Ingo FrA 1/4nd, Dr. rer. nat.
>      Modelling of Cognitive Processes
>      TU-Berlin
>      Sekr 6-4
>      Franklinstrasse 28/29
>      10587 Berlin
> 
>      phone: (+49|0)30-314-78962
>      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Ingo Fründ, Dr. rer. nat.
Modelling of Cognitive Processes
TU-Berlin
Sekr 6-4
Franklinstrasse 28/29
10587 Berlin

phone: (+49|0)30-314-78962
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~