hddm-users Google Group

Re: [hddm-users] hddm instalation problem

thomas.wie...@gmail.com (Thomas Wiecki) — Sun, 19 May 2013 19:44:08 UT

Did you install cython?

Thomas

Re: [hddm-users] hddm instalation problem

andrecr...@gmail.com (Andre Cravo) — Sun, 19 May 2013 18:39:50 UT

Hi Thomas,

Still not working. I'm sorry for all the trouble, here is the new log. Any
ideas what might be wrong?

Best

Re: [hddm-users] hddm instalation problem

thomas.wie...@gmail.com (Thomas Wiecki) — Fri, 17 May 2013 12:03:42 UT

This same error happened with python x,y:
[link]

Re: [hddm-users] hddm instalation problem

thomas.wie...@gmail.com (Thomas Wiecki) — Fri, 17 May 2013 12:01:49 UT

This is the problem:
build\temp.win32-2.7\Release\s rc\cdfdif_wrapper.o:cdfdif_wra pper.c:(.text+0x6133):
undefined reference to `imp(char, double, float, double, int, float)'

It might be a Cython problem, can you try installing that and then
reinstall hddm?

Thomas

Re: [hddm-users] hddm instalation problem

andrecr...@gmail.com (Andre Cravo) — Fri, 17 May 2013 01:22:47 UT

Hi Thomas,

Thanks for your help. I have tried to use Anaconda and it solves part of
the problem. But now I have a different problem. I have been able to
install all other packages (kabuki, pandas and all). However, when I try:
pip install hddm I get:

Command C:\Anaconda\python.exe -c "import

Re: [hddm-users] hddm instalation problem

thomas.wie...@gmail.com (Thomas Wiecki) — Thu, 16 May 2013 11:35:18 UT

It seems you can also locate distutils.cfg and change the compile to
mingw32:

I had a similar problem with a package (Traits) a couple of weeks ago - for
me it was because the package was trying to compile extensions and I didn't
have Visual Studio. What worked for me was to install
MinGW<[link]> and

Re: [hddm-users] hddm instalation problem

thomas.wie...@gmail.com (Thomas Wiecki) — Thu, 16 May 2013 11:30:05 UT

Hi Andre,

That's rather odd, EPD should use mingw, not visual studio. You can try to
install Visual Studio (there's a free version) manually and see whether
that works. Alternatively you might want to try Anaconda by Continuum
instead of EPD.

Anyone else with windows experience that knows anything?

hddm instalation problem

andrecr...@gmail.com — Thu, 16 May 2013 01:48:28 UT

Hello,
I am trying to install hddm on windows 7 (32-Bits). I have downloaded and
installed Enthought Python Distribution (EPD).
I have also been able to install kabuki.
However, when I try to install hddm I get the following error:
error: Unable to find vcvarsall.bat
I am sending the full log attached. Would somebody know the solution to

Re: [hddm-users] Quantile Optimization

dunov...@gmail.com (DunovanK) — Wed, 15 May 2013 20:59:56 UT

Imri, Michael, and EJ,
Thank you for the feedback. I know that BIC is known to overpenalize, but
also that DIC can have the opposite problem. Thus, using a more
conservative criterion measure (like BIC) to compare models seemed like a
nice way to ensure that more complex models were truly fitting the data

Re: [hddm-users] Quantile Optimization

michael_fr...@brown.edu (Michael J Frank) — Wed, 15 May 2013 17:35:23 UT

EJ, I defer to you on these issues. (I've been thinking a lot about model
selection and bayes factors etc lately so I hope to be able to chat with
you about this when in NL in June!)

My understanding of the Stephan paper was that rather than assuming that
there is one model that generates the data for all subjects (as in group BF

Re: [hddm-users] Quantile Optimization

ej.wagenmak...@gmail.com (Eric-Jan Wagenmakers) — Wed, 15 May 2013 15:20:37 UT

My two cents: I never understood the Stephan paper. At best it tells you
something about the average evidence across subjects. However, what one
likes to know is the overall evidence. If every subject contributes a
little evidence for H1, and you have many subjects, you need to end up with
a lot of evidence for H1. Evidence tends to sum or multiply, not average.

Re: [hddm-users] Quantile Optimization

michael_fr...@brown.edu (Michael J Frank) — Wed, 15 May 2013 14:49:21 UT

I agree with Imri about BIC, but just add one bit about your question re:
combining individual BIC scores to obtain a group level BIC. A nice paper
for this is

Bayesian model selection for group studies.
Stephan KE, Penny WD, Daunizeau J, Moran RJ, Friston KJ.
Neuroimage. 2009 Jul 15;46(4):1004-17.

Re: [hddm-users] Quantile Optimization

imriso...@gmail.com (Imri Sofer) — Wed, 15 May 2013 12:37:02 UT

Hi Kyle,
honestly I don't remember the whole theory behind the optimization scores,
just that their is a close connection between G^2 and BIC. So I cannot
fully answer your question.

I can tell you that you should take the BIC score with a huge chunk of salt.
BIC complexity penalty just count the number of parameters, it does not

Re: [hddm-users] Error in saving model

mmasaki1...@gmail.com (Maruyama Masaki) — Wed, 15 May 2013 12:28:08 UT

Hello Thomas,

I always appreciate your kind advices.

I have done the two trials.
1. I can save&load without any error or warning message.
2. I can save with warning message "WARNING: Will not save custom link
functions." and can load with warning message "WARNING: Custom link
functions will not be loaded.".

Re: [hddm-users] Quantile Optimization

dunov...@gmail.com (DunovanK) — Wed, 15 May 2013 00:30:10 UT

Hi Thomas,
I'm finding the quantile optimization to be a very useful tool for testing
out potential model configurations before fitting them to the data with
Hierarchical MCMC (which, as you've mentioned, takes a much longer time to
estimate). In tinkering with the various fitting methods (X2, G2, MLE),