Classification with SVM

[ThorPedo]

Hi - many thanks for an excellent machine learning library.
I have been testing the binary classification sample and extending it to a vector of 12 parameters (the sample has two, x and Y) . It works, but if I run the training set the return values are of course exact (1 and -1 in correct places), but if I run a large dataset which also includes the values of the training set it does not even detect the subset which is the training set. Shouldn't it at least do that? Am I missing something here? the attached picture shows screen shot of the training data and results from two runs with the training set as input and a large data set as input  The training data has 1000 items (each 12 parameters) and the large data set has 124000 items.  why am I not getting a working classification out of this trial.