unixjunkie/orsvm-e1071

Access the Linear or RBF kernel SVM from OCaml using the R e1071 or svmpath packages

orsvm-e1071

Access from OCaml SVMs provided by the R e1071 and svmpath packages.

OCaml functions to drive the RBF and linear kernel SVMs of the R e1071 package and the linear kernel SVM of the R svmpath package.

This is low level on purpose: we don't provide access to all functionalities offered by the R packages.

This is probably not high performance: data are exchanged between R and OCaml via ASCII files.

orsvm-e1071's Issues

we should support the radial kernel in svmpath too

add svmpath.mli

tmp files are in a limited number unless we use long XXX prefixes

print AUC on training set in test.ml

release in opam

add debug flag to both train and predict

on debug (false by default): log the temp R script filename and don't delete it before exit

make new release once the fst one is OK in opam

automatic optimization of cost and gamma using N folds cross validation on the training dataset

target function: optimization of AUC

get rid of preprocessor

this is uggly and forces me to do some jbuilder programming via its jbuild file.
I have never programmed ocamlbuild, let's do the same with jbuilder.

Error on SVR cross-validation using tune

Hi there,

I am using the function tune to fit a support vector regression using a cross-validation. The independent variable is a sparse document-term matrix and I am trying to fit them on stock market and bond market returns.

When I run the cross-validation:

tune(svm, train_data, labels,
ranges = list(epsilon = seq(0,1,0.1), cost = 2^(2:9)))

I get following error:

Error in h(simpleError(msg, call)) :
error in evaluating the argument 'x' in selecting a method for function 't': error in evaluating the argument 'x' in selecting a method for function 't': Subscripts out of bound

This error depends on the label that I use (sometimes it works and sometimes in doesn’t). Do you have an idea where does this come from? I could provide an example if needed.

Thanks!

support sparse data

both the R package and libsvm (that it uses underneath) support that

• we need to know how to store such data in a txt file
• we need to know how to tell R to read in such a file

add .mli file with coments on expected file formats

expose the linear kernel in some way

we'll have two flavors then: RBF | Linear

read_predictions: rm file upon success and if not debug