Non-sparse formats

[beevabeeva]

Hi,

I have been planning to use ThunderSVM (and ThunderGBM) in my research. Briefly, this research consists of implementing state of the art GPU algorithms in the ATM framework. The ATM framework uses non-sparse matrix as the input format. I have posted an issue regarding this on their repo too. I know that your work is based on LIBSVM and that the input format is somewhat of a contentious issue. But my question is: How difficult would it be to implement support for non-sparse/LIBSVM input formats? In particular, ATM expects input in the form:

feature_01,feature_02,feature_03,feature_04,class
5.1,3.5,1.4,0.2,Iris-setosa
4.9,3.0,1.4,0.2,Iris-setosa
4.7,3.2,1.3,0.2,Iris-setosa
4.6,3.1,1.5,0.2,Iris-setosa

• It uses a single comma, ,, as the separator.
• Its first row is a header that contains the names of the columns.
• There is a column that contains the target variable that will need to be predicted.
• The rest of the columns are all variables or features that will be used to predict the target column.
• Each row corresponds to a single, complete, training sample. (see this page )

Thanks,

Avi

[zeyiwen]

A parser for the data format you have mentioned will be needed. You are more than welcome to contribute.