Are SVM/RF implementations available through scmap?

[CodeInTheSkies]

Hi there,

The help page of of scmapCluster mentions that we can potentially give the probability values for the SVM/RF classifiers through the "threshold" argument. There are also such functions for SVM and RF in the "Utils.R" script on scmap's repository here: https://github.com/hemberg-lab/scmap/blob/92566d3db92d5595eedbdf20c7f03aa2a1a053ac/R/Utils.R

So, can we actually get scmap to implement the SVM and RF methods of classifications that you actually use in the paper for any given dataset?

Just wondering, and would appreciate your answer!

Thanks!

[wikiselev]

Hi, yes, the functions that we used for SVM and RF are still in the utils. The inputs to these functions would be just expression matrices:

scmap/R/Utils.R

Line 114 in fd54f6b

support_vector_machines <- function(train, study, kern = "linear") {

Calls to these functions were used in the old versions of scmap, if you browse the github history:

scmap/R/CoreMethods.R

Line 266 in b3ef5b6

if (method == "svm") {

Unfortunately, I removed them from main function at some point in the past and not sure that if you add them now, they will work (some of the functions inside these if blocks might have been also removed). However, if there is a demand of them from the users we could probably add them in the future versions, but this is not a priority at the moment.

[CodeInTheSkies]

OK, thanks! For addition of these functions in the future, do count me in as one vote!