[pymvpa] Searchlight sl_map = accuracy?

Nynke van der Laan nynkevanderlaan at gmail.com
Tue Feb 22 10:13:17 UTC 2011


Dear Yaroslav,
thanks for your quick reply. Just to have things clear: the definition
of TransferError refers to the accuracy of the classifier's
predictions on a novel dataset (first alinea of chapter 5.2 Error
calculation). So the CrossvalidatedTransferError is the mean of the
transfererrors (i.e., accuracies) over the validations. From your
reply I understand that the outcomes of the analysis in the example
(the Searchlight on fMRI data) are not the accuracies but the inverse
(i.e., 1 - accuracy). Would you please confirm if this is right?
Thanks for the example with the conversion to Nifti, that helped me a lot!
Best regards,
Nynke

On Mon, 21 Feb 2011, Nynke van der Laan wrote:
> First, what do the values in the sl_map refer to? According to the
> documentation, these are the TransferErrors of the
> n-foldclassification.
yeap

or extended: output of Searchlight corresponds to the measure you
provide, in this case it is CrossValidatedTransferError, which in turn
aggregates TransferErrors with default argument
errorfx=MeanMismatchErrorFx();  thus values are indeed errors in your
case

> Later on it is called an sensitivitymap. But
> does this mean that it are the accuracies? (true pos + true neg/
> total) or are it sensitivities? (true positive/all positives)

sorry for not-clear variable naming I guess -- there is no magic
conversion happening from above errors into accuracies happening -- it remains
the same error(s) (i.e. 1.0 - accuracy = 1.0 - (TP+TN)/total)

> Second, I read that it is possible to map the values from the sl_map
> back to the original dataspace. In the example the variable
> masked_orig_sl_map is the sensitivity map that is mapped back into the
> original dataspace. How can I save this variable as a nifti file in
> order to be able to use it as an 3d-overlay for in, for example,
> MRIcron or another imageviewer?

please look at another example, which haven't made into the compiled
manual:

https://github.com/PyMVPA/PyMVPA/blob/maint%2F0.4/doc/examples/searchlight.py#L112

the critical function for that is map2Nifti

hope this helps
-- 
=------------------------------------------------------------------=
Keep in touch                                     www.onerussian.com
Yaroslav Halchenko                 www.ohloh.net/accounts/yarikoptic



More information about the Pkg-ExpPsy-PyMVPA mailing list