svmtrain

Hello all,

I have two sets of data each of size 3x400. I would like to train them using svmtrain but I don't know what I should put for group. Can you please help me? can you also let me know what should I put for group if I get data with arbitrary sizes? thanks.

Answer 1

[Walter Roberson]

Each data sample must belong to a pre-known group for the purpose of svmtrain. The grouping variable is a list of which group each sample belongs to.

Do you have 3 features or 400 features? Are your two data sets indicating exactly two different groups, or are there entries from multiple groups within a data set? If your data is already split by groups, then your group variable should be 1 for each member of the first data set and 2 for each member of the second data set.

[Bahareh]

I have 3 features with 400 observations and each data sets corresponds to one class. so you mean I should use svmtrain for each class separately? i.e. can't I put all my data in one array with size 3x800 and train them?

[Walter Roberson]

Approximately,

svm( [setA,setB].', [ones(size(setA,2),1);2*ones(size(setB,2),1)] )

That is, the data submitted to svm must be one observation per _row_, so splice the data together and switch from column-oriented to row-oriented. The group variable will be 1 for all of the observations in setA and 2 for all of the observations in setB.