fitEnsembleModel: fitEnsembleModel

In huangwb8/GSClassifier: Gene Signature Classifier

fitEnsembleModel

Description

Train multiple subtype models using cross validation

Usage

fitEnsembleModel(
  Xs,
  Ys,
  geneSet = NULL,
  na.fill.method = c("quantile", "rpart", NULL)[1],
  na.fill.seed = 2022,
  n = 20,
  sampSize = 0.7,
  sampSeed = 2020,
  breakVec = c(0, 0.25, 0.5, 0.75, 1),
  params = list(device = "cpu", nrounds = 15, max_depth = 10, eta = 0.5, nthread = 5,
    colsample_bytree = 1, min_child_weight = 1),
  nround.mode = c("fixed", "polling")[2],
  xgboost.seed = 105,
  caret.grid = expand.grid(nrounds = c(10, 15), max_depth = c(5, 10), eta = c(0.01, 0.1,
    0.3), gamma = c(0.5, 0.3), colsample_bytree = 1, min_child_weight = 1, subsample =
    0.7),
  caret.seed = 101,
  ptail = 0.5,
  verbose = F,
  numCores = 2
)

Arguments

Xs	Gene expression matrix.
Ys	Phenotype vector, multiclass
geneSet	A list of genes for classification
na.fill.method	Missing value imputation method for na_fill function. One of 'quantile', 'rpart' and NULL.
na.fill.seed	Seed for na_fill function.
n	Size of the ensember, where each member is a result from fitSubtypeModel
sampSize	proportion of samples to hold back
sampSeed	random seed for subset of Xs
breakVec	vector of break points, used to bin expression data
params	The parameters for xgb.train. 1. xgb.cv only: nfold; 2. xgboost: nrounds, max_depth, eta, nthread, colsample_bytree, min_child_weight.
nround.mode	One of fixed and polling. fixed mode is recommended! polling Default but legacy feature, which means to call the best_iteration via xgb.cv fixed Use the default nrounds in params, so it's faster(10-20 folds) than polling.
xgboost.seed	Seed for xgboost.
caret.seed	The random seed for caret::train process when params is NULL
ptail	Binary phenotype vector.
verbose	whether report modeling process

Details

The geneid of geneSet and Xs must be the same (one of ENSEMBL, SYMBOL or ENTREZID). In addition, if fitEnsembleModel is hanged on, please check the space use of /, which would disturb the work of makeCluster.

Value

A list of lists of xgboost classifiers

huangwb8/GSClassifier documentation built on July 12, 2024, 5:10 p.m.