syntax questions

[Inge Leunissen]

Hi, 

First of all congratulations on the update, everything works smoothly so far! 

I've got 2 naive syntax questions. 

1) I've ran a group model with a condition effect on 'a' and I'm trying to compute the difference between conditions by means of the group_cond_diff function I found in kabuki.analyze

>>> group_cond_diff(model,'a','repeat','switch')

Traceback (most recent call last):

  File "<stdin>", line 1, in <module>

  File "C:\Python27\lib\site-packages\kabuki\analyze.py", line 215, in group_con

d_diff

    node_dict = hm.params_include[name].subj_nodes

AttributeError: 'HDDM' object has no attribute 'params_include'

I assume that I did not enter the node name correctly but I've tried a number of possibilities and it keeps giving the same error message. Can anyone help me out with the correct syntax? 

Moreover, I've also got a model which contains a group*condition interaction, any tips on how I can adjust the script such that it loops over subjects and groups?

2) I tested convergenge by means of the posterior plots and geweke. I would however like to run the gelman_rubin. Since I've got a group model the MAP method cannot be used and it is suggested to use the subj_by_subj_map_init. I cannot seem to find this function anywere, how should it be called? 

Thanks a lot! 

Inge 

[Inge Leunissen]

Sorry, I've been reading some of the previous post and just to make things clear: the model is hierarchical, and not a 'flat' group model 

[Thomas Wiecki]

Hi Inge,

On Wed, Sep 12, 2012 at 6:48 AM, Inge Leunissen <ingele...@gmail.com> wrote:

Hi, 

First of all congratulations on the update, everything works smoothly so far! 

I've got 2 naive syntax questions. 

1) I've ran a group model with a condition effect on 'a' and I'm trying to compute the difference between conditions by means of the group_cond_diff function I found in kabuki.analyze

>>> group_cond_diff(model,'a','repeat','switch')

Traceback (most recent call last):

  File "<stdin>", line 1, in <module>

  File "C:\Python27\lib\site-packages\kabuki\analyze.py", line 215, in group_con

d_diff

    node_dict = hm.params_include[name].subj_nodes

AttributeError: 'HDDM' object has no attribute 'params_include'

I assume that I did not enter the node name correctly but I've tried a number of possibilities and it keeps giving the same error message. Can anyone help me out with the correct syntax? 

Unfortunately it seems this function is deprecated (we missed that somehow). However, it is quite easy to compute the difference between two (group) nodes manually:

diff_trace = model.nodes_db['a_cond1'].node.trace() - model.nodes_db['a_cond2'].node.trace()
diff_mean = numpy.mean(diff_trace)
p_sig = numpy.mean(diff_trace < 0) # check for prob mass < 0

Moreover, I've also got a model which contains a group*condition interaction, any tips on how I can adjust the script such that it loops over subjects and groups?

Checking if the group mean differs might be good enough. But in general you can access each node individually via the nodes_db.

2) I tested convergenge by means of the posterior plots and geweke. I would however like to run the gelman_rubin. Since I've got a group model the MAP method cannot be used and it is suggested to use the subj_by_subj_map_init. I cannot seem to find this function anywere, how should it be called? 

Not sure why you need the MAP for gelman_rubin?

subj_by_subj_map_init has been changed to: model.approximate_map() 

Thomas

Thanks a lot! 

Inge