In aydindemircioglu/SVMBridge: A Bridge to Integrate Different 'SVM' Solvers into R

Writing Wrappers

To write your own wrapper, you simply have to provide an R-file that handles the most important things like reading and writing models, and communicating with the executables. Technically, a wrapper is just an S3 Object. We will talk about the methods below.

Notice that if you write your own findSoftware routine for your wrapper, you might need some platform dependent checks, e.g. the LIBSVM binray is simply called "svm-train" on Linux and Mac OSX, but "svm-train.exe" on Windows. You can use the functionality provided by R to check for the platform. An example can be seen in the LIBSVM wrapper code.

S3 Object

A wrapper is a simple S3 Object. If you do not know about S3 Object, you might similarly think about providing several callback functions to the SVMBridge. The most complex part will be provide functions to read and write the model. As a starter, you might want to look into the prepackaged wrappers, SVMBridge provides.

The following parts have to be written, where x is the S3 Object.

createTrainingArguments

createTrainingArguments = function (x, trainDataFile = NULL, modelFile = NULL, verbose = FALSE, ...)
  trainArgs = ...
  return (trainArgs)

This function will transform given parameters into a string that will be used as arguments when executing the corresponding SVM train binary. It will return the string.

Note: Please note that as the training arguments is passed as a string, it might need proper escaping. This can be done by using shQutoes, as you can see in the prepackaged wrappers

createTestArguments

createTestArguments = function (x, testDataFile = NULL, modelFile = NULL, predictionsFile = NULL, verbose = FALSE, ...) {
  trainArgs = ...
  return (trainArgs)

This function will transform given parameters into a string that will be used as arguments when executing the corresponding SVM test binary. It will return the string.

Note: Please note that as the training arguments is passed as a string, it might need proper escaping. This can be done by using shQutoes, as you can see in the prepackaged wrappers

extractTrainInfo

extractTrainInfo = function (x, output, verbose = FALSE) {
    trainList = ...
  return (trainList)
}

This function will extract information from the output of the SVM train binary. In general, either a list or a scalar can be returned. In case of a scalar, this must be the training error. If it is a list, then training error should be part of the this list.

extractTestInfo

extractTestInfo = function (x, output, verbose = FALSE) {
  testList = ...
  return (testList)
}

This function will extract information from the output of the SVM test binary. In general, either a list or a scalar can be returned. In case of a scalar, this must be the test error. If it is a list, then test error should be part of the this list.

readModel

readModel = function (x, modelFile = './model', verbose = FALSE) {
  model = ...
  return (model)
}

This function will read a model on disk into memory. The model should be stored as a list and returned.

writeModel

writeModel = function (x, model = NA, modelFile = "./model", verbose = FALSE) {
  err = ...
  return (err)
}

This function will write a model to disk. It will return an error code.

detectModel

detectModel = function (x, modelFile = "./model", verbose = FALSE) {
    return (bool)
}

This function will determine if a given model on disk is compatible with the model for the SVM solver. Usually a simple check (e.g. does the file start with the line 'SVM') is conducted. It will return either TRUE or FALSE.

readPredictions

readPredictions = function (x, predictionsFile = NULL, verbose = FALSE) {
    return (predictions)
}

This function will read the predictions from the predictions file and return them as a list.

findSoftware

findSoftware = function (x, searchPath = "./", verbose = FALSE) {
    x$trainBinaryPath = ...
    x$testBinaryPath = ...
    return (x)
}

This function will determine whether the binaries of the SVM solver exist in the given searchPath or not. It shall not do any kind of recursive search (this is done in the SVMBridge itself), but only seek out for the test and trining binary. Both should be put back into the object x and x must be returned. If the binaries are not found, NULL should be put into the variables instead.

aydindemircioglu/SVMBridge documentation built on May 11, 2019, 4:13 p.m.