[pymvpa] Sensitivity analysis also fails

[Temilade Adelore]

Hello Yaroslav,
Thank you for your response!

I am new to MVPA and so I am not clear how else I should correctly define
'chunks' for event related data. Here's the scenario:

I am trying to work out how to correctly code the chunks for:
1) An event related data with conditions within each chunk (10) presented
3.5 seconds in duration 2) and volume samples of 2 seconds duration

Thus a Nifti dataset with its number of samples (1315) greater than its
number of chunks (609).

I am having a problem coding this in because pymvpa likes both number of
sample and number of chunks to be equal.

How do you suggest I code in the chunks for this dataset?

Also I have managed to compound all Nifti images across all runs into one 4D
data.
In trying to move forward, I loaded the data as event related but did not
explicitly specify the number of chunks. However, when I try to detrend ,
zscore or classify the data I get: Memory Error

The script for reference is at http://gist.github.com/367912
Any help will be greatly appreciated!

Thank you.

Temi



On Thu, Apr 15, 2010 at 10:22 PM, Temilade Adelore <temiadelore at gmail.com>wrote:

> Hello Yaroslav,
>
> Thank you for responding.
> Here the link to my code on gist <http://gist.github.com/367912>
>
> At line 93 in the script I get the following error:
>
> File "/usr/lib/pymodules/python2.6/matplotlib/image.py", line 294, in
> set_data
> raise TypeError("Invalid dimensions for image data")
> TypeError: Invalid dimensions for image data
>
> At line 116 in the script I also get the following error:
>
> File "/usr/lib/pymodules/python2.6/mvpa/clfs/base.py", line 369, in train
> if dataset.nfeatures == 0 or dataset.nsamples == 0:
>
> AttributeError: 'NoneType' object has no attribute 'nfeatures'
>
> However when I type in following in the shell I get:
>
> >>> fmri_src_ev.nfeatures
> 405504
> >>> fmri_src_ev.nsamples
> 44
>
> And so it is clear that my dataset (fmir_src_ev) has the nfeatures and
> nsamples attribute but this is not being recognized by the sensitivity
> analyzer/classifier.
>
> I am new to pymvpa and so I understand that I might be doing something
> wrong.
> Any help or suggestions would be greatly appreciated.
>
> Thanks,
> Temi Adelore
>
>
>
> On Thu, Apr 8, 2010 at 4:55 PM, Temilade Adelore <temiadelore at gmail.com>wrote:
>
>> Hello Experts,
>>
>> when I use run the following code:
>>
>> clf = LinearCSVMC()
>>
>> sclf = SplitClassifier(clf, NFoldSplitter(),
>>        enable_states=['confusion', 'training_confusion'])
>>
>> sensitivities = \
>> sclf.getSensitivityAnalyzer(combiner=None,
>>
>> slave_combiner=None)(fmri_src_ev)
>>
>> I get the error: AttributeError: 'NoneType' object has no attribute
>> 'nfeatures'
>>
>> However when I type in "fmri_src_ev.nfeatures" in the terminal
>> I get back: "405504"
>>
>> And so I am confused as to what might be the problem.
>>
>> Pls help!! Any feedback would be greatly appreciated.
>>
>> Thanks, Temi
>>
>>
>>
>>
>> On Thu, Apr 8, 2010 at 4:22 PM, Temilade Adelore <temiadelore at gmail.com>wrote:
>>
>>> Hello Experts,
>>> I am trying to load samples into pymvpa using the following command:
>>>
>>> pre_fmri_src = NiftiDataset(samples=[fmri_src1, fmri_src2, fmri_src3,
>>> fmri_src4, fmri_src5, fmri_src6, fmri_src7, fmri_src8,
>>> fmri_src9,fmri_src10], labels=4, chunks=10, enforce_dim=None, mask=None)
>>>
>>> Each of the 10 samples correspond to a 4D NiftiImage instance.
>>> I would like each sample to be a chunk, hence creating 10 chunks of data.
>>> However, when I run the command above I get the following error:
>>>
>>> ValueError: Input volumes contain variable number of dimensions: [(136,
>>> 33, 64, 64), (136, 33, 64, 64), (136, 33, 64, 64), (136, 33, 64, 64), (136,
>>> 33, 64, 64), (136, 33, 64, 64), (136, 33, 64, 64), (136, 33, 64, 64), (136,
>>> 33, 64, 64), (91, 33, 64, 64)]
>>>
>>> I realize that the last chunk has a different dimension ( a different
>>> number of time points) however I do not see why should give me an error.
>>> Is there anyway I can work around this? All suggestions are much
>>> appreciated.
>>> Thanks,
>>> Temi Adelore
>>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/pkg-exppsy-pymvpa/attachments/20100422/aafcab8e/attachment.htm>