LedPred: Creates an SVM model given a feature matrix

In LedPred: Learning from DNA to Predict Enhancers

Description

The LedPred function computes the best SVM parameters, defines the optimal features for creating the SVM model by running sequentially mcTune, rankFeatures, tuneFeatureNb and createModel. The performances of this model are then computed usong evaluateModelPerformance.

Usage

LedPred(data = NULL, cl = 1, ranges = list(gamma = c(1, 10), cost = c(1,
  10)), cost = NULL, gamma = NULL, kernel = "linear", valid.times = 10,
  file.prefix = NULL, numcores = ifelse(.Platform$OS.type == "windows", 1,
  parallel::detectCores() - 1), step.nb = 10, halve.above = 100)

Arguments

data	data.frame containing the training set
cl	integer indicating the column number corresponding to the response vector that classify positive and negative regions (default = 1)
ranges	list object containing one (linear kernel) or two (radial kernel) vectors of integers corresponding to SVM cost and SVM gamma parameters to test.
cost	The SVM cost parameter for both linear and radial kernels. If NULL (default), the function mcTune is run.
gamma	The SVM gamma parameter for radial kernel. If radial kernel and NULL (default), the function mcTune is run.
kernel	SVM kernel, a character string: "linear" or "radial". (default = "radial")
valid.times	Integer indicating how many times the training set will be split for the cross validation step (default = 10). This number must be smaller than positive and negative sets sizes.
file.prefix	A character string that will be used as a prefix for the result files. If it is NULL (default), no plot is returned
numcores	Number of cores to use for parallel computing (default: the number of available cores in the machine - 1)
step.nb	Number of features to add at each step (default = 10)
halve.above	During RFE, all the features are ranked at the first round and the half lowest ranked features (that contribute the least in the model) are removed for the next round. When the number of feauture is lower or equal to halve.above, the features are removed one by one. (default=100)

Value

A list of the object produced at each step

best.params	A list of the parameters giving the lowest misclassification error
feature.ranking	List of ordered features from rankFeatures
feature.nb	he optimal number of feature to use from the list of ordered features from tuneFeatureNb
model.svm	The best SVM model createModel
probs.label.list	The cross-validation results from evaluateModelPerformance

Examples

 data(crm.features)
 #cost_vector <- c(1,3,10)
 #gamma_vector <- c(1,3,10)
 #ledpred.list=LedPred(data.granges=crm.features, cl=1, ranges = list(cost=cost_vector,
 #                          gamma=gamma_vector), kernel="linear", halve.above=50)
 #names(ledpred.list)

LedPred documentation built on Nov. 8, 2020, 8 p.m.