Type results=Experiments('USPST.mat') etc to see the
results for some datasets (COIL20.mat, TEXT1.mat G50C.mat, USPST.mat
files) except the WEBKB dataset (since we used a different protocol
for WebKB as it has very few labeled examples). 

Type results=Experiments('USPS_test') to see transduction and
out-of-sample results for a USPS problem with 1607 labeled and
unlabeled examples, and 400 unseen test examples. 

results is a struct array containing fields for each split. It records
test and transduction confusion matrix, error rate and also the real
valued outputs for both svm and rlsc.

The semi-supervised kernel is implemented in Deform.m which takes a
kernel matrix and a graph regularizer as inputs, and outputs the
semi-supervised kernel.

Type ExperimentsWebKB('page') etc to see results on the WebKB dataset
('page' for page view, 'link' for link view and 'page+link' for
page+link view).

Type ExperimentsWebKB('page',1) etc for the results using a multi-view
regularizer.  The second argument tells the code to use the mean
laplacian in the 3 views to learn in each (These are results labeled
LapSVM_joint, LapRLS_joint in the table).

The SVM is implemented thru mex files so you would have to compile the
mex file :

mex -c svmprecomputed.cpp mex 
mexGramSVMTrain.cpp svmprecomputed.o

Some files that may be useful as stand-alone code:

laplacian.m : Computes graph Laplacian. 
calckernel.m: Computes standard kernels.

Please email me at vikass@cs.uchicago.edu for help/questions about this 
code.
