<div dir="ltr"><div class="gmail_default" style="font-family:georgia,serif;font-size:large;color:rgb(0,0,0)">Hi,<br><br>Thank you for the great software. I am really enjoying it.<br></div><div class="gmail_default" style="font-family:georgia,serif;font-size:large;color:rgb(0,0,0)">We have been learning about the searchlight analyses<br></div><div class="gmail_default" style="font-family:georgia,serif;font-size:large;color:rgb(0,0,0)"> as part of a broader analyses, and have encountered an issue for which we have no apparent explanation. The univariate GLM substraction analyses show strong differences in activation in a frontoparietal network betwen a *cued* and *uncued* attention task. We performed a searchlight analyses with KNN and SVM classifiers on the same data *as a sanity check*<br clear="all"></div><div class="gmail_default" style="font-family:georgia,serif;font-size:large;color:rgb(0,0,0);display:inline">expecting to find spheres in frontoparietal cortex</div><div class="gmail_default" style="font-family:georgia,serif;font-size:large;color:rgb(0,0,0);display:inline"> with significant classification accuracy. But we do not find anything like it.<br><br></div><div class="gmail_default" style="font-family:georgia,serif;font-size:large;color:rgb(0,0,0);display:inline">There are two experimental conditions *cued and uncued* and 19 subjects. <br><br>The data input to the searchlight is the parameter estimates (PEs) from the GLM that give such strong univariate effecs in frontoparietal cortex</div><div class="gmail_default" style="font-family:georgia,serif;font-size:large;color:rgb(0,0,0);display:inline">.<br><br></div><div class="gmail_default" style="font-family:georgia,serif;font-size:large;color:rgb(0,0,0);display:inline">We therefore have a 4D nii file in which volumes 1-19 are PEs for the *cue* classification target and volumes 20-38 are the PEs for the *uncued* classification target<br><br></div><div class="gmail_default" style="font-family:georgia,serif;font-size:large;color:rgb(0,0,0);display:inline">This is the code we use. In it the chunks are the subject (1-19)<br><br>from mvpa2.suite import * <br>attr = SampleAttributes(os.path.join(datapath, 'attr1task_cued_vs_uncued.txt'))<br>ds = fmri_dataset(samples=os.path.join(datapath1, 'introsp_pred.nii.gz'),<br> targets=attr.targets, chunks=attr.chunks,<br> mask=os.path.join(datapath2, 'mask.nii.gz'))#based on FEAT analyses<br> #add_fa={'unmbral_glm': os.path.join(datapath2, 'mask.nii.gz')})<br>#<a href="http://ds.sa">ds.sa</a>['task']=np.repeat(['search','introsp'], 16) <br>zscore(ds, chunks_attr='chunks')<br><br>clf=kNN(k=1, dfx=one_minus_correlation, voting='majority') <br>#clf = LinearNuSVMC()<br>partitioner=NFoldPartitioner() <br>cv=CrossValidation(clf, partitioner)<br><br>sl=sphere_searchlight(cv, radius=3, space='voxel_indices', postproc=mean_sample(), nproc=4)<br>res_knn=sl(ds)<br>niftires = map2nifti(ds, res_knn)<br>archivo='/home/dsoto/Documents/fmri/wmintrosp/just1indivcopesML/orig/sl_knn.nii.gz' <br>niftires.to_filename(archivo)<br>h5save(os.path.join(datapath1,'sl_knn.hdf5'),res_knn_pred_z_by_subj) <br> <br></div><div class="gmail_default" style="font-family:georgia,serif;font-size:large;color:rgb(0,0,0);display:inline">The above code gave a Warning re: the zscoring stage becos the number of datapoints per chunks was only 2 (in the example above the chunks were the subjects)<br><br></div><div class="gmail_default" style="font-family:georgia,serif;font-size:large;color:rgb(0,0,0);display:inline">I ALSO NOTE that I repeated this using different zscoring approaches (eg zscoring by each subject, by each condition *cued* vs *uncued* or by the whole input data, for this later we created a column factor with *1* for the 38 data points)<br></div><div class="gmail_default" style="font-family:georgia,serif;font-size:large;color:rgb(0,0,0);display:inline"></div><br><br><div class="gmail_default" style="font-family:georgia,serif;font-size:large;color:rgb(0,0,0);display:inline">I am confused as to why the univariate GLM gives such strong difference between conditions, while the searchlight gives null classification accuracy.<br><br></div><div class="gmail_default" style="font-family:georgia,serif;font-size:large;color:rgb(0,0,0);display:inline">Any input is highly appreciated!<br><br>cheers<br>ds</div><br>-- <br><div dir="ltr"><br><a href="http://www1.imperial.ac.uk/medicine/people/d.soto/" style="color:rgb(17,85,204)" target="_blank">http://www1.imperial.ac.uk/medicine/people/d.soto/</a><br></div>
</div>