kentman234 Wrote:
-------------------------------------------------------
> I want to do a project about classification of
> fMRI data. I have data for 20 subjects who see two
> types of videos. The data have been already
> preprocessed
How exactly? Do you have different predictors for the two types of movies in a GLM?
Multiple exemplars for each type or just one? Are you using all voxels, or are you considering a region of interest?
> and my idea is to train (10 subjects
> for example) and test the other 10 subjects
Between participant classification - that sounds challenging.
> to calculate the prediction accuracy and
> estimating the spatial maps from SVM lineary
> weight vector. I want you to help me with 3dsvm
> program as i'm new to using it, for example when i
> write the command: (3dsvm - trainvol run1+orig\ )
> i have a fatal error: must specify timeseries
> labelfile for training!
I don't have experience with 3dsvm, but what I understand is that if you have a training set with N subbriks (these could either be form different exemplars of one participant, or combined data from multiple participants), then you need a 1D file with N values that contain the training labels. To make your training set you can join and select subbricks from different files using 3dTcat or 3dbucket.
Did you read the help of 3dsvm (which you get by running 3dsvm without arguments)? In the help the first example shows how to specify the training labels with -trainlabels.
Again, your project seems quite challenging, and you might consider starting simple and making sure you get 'simple' within-participant classification working first.
Hope this helps,
nick