ks.benchmark: ks.benchmark

In kstawiski/miRNAselector: miRNAselector - a package for biomarker selection based on high-throughput experiments.

Description

Second most important function in the package. Using the formulas selected by 'ks.miRNAselector' function, it test derived miRNA sets in a systematic manner using multiple model induction methods. This function allows to benchmark miRNA sets in context of their potential for diagnostic test creation. Hidden feature of this package is application of 'mxnet'. Note that 'mxnet' has to be installed and configured seperatly.

Usage

ks.benchmark(
  wd = getwd(),
  search_iters = 2000,
  keras_epochs = 5000,
  keras_threads = floor(parallel::detectCores()/2),
  search_iters_mxnet = 5000,
  cores = detectCores() - 1,
  input_formulas = readRDS("featureselection_formulas_final.RDS"),
  output_file = "benchmark.csv",
  mxnet = F,
  gpu = F,
  algorithms = c("mlp", "mlpML", "svmRadial", "svmLinear", "rf", "C5.0", "rpart",
    "rpart2", "ctree"),
  holdout = T,
  stamp = as.character(as.numeric(Sys.time()))
)

Arguments

wd	Working directory here 'ks.miRNAselector' was also working.
search_iters	The number of random hyperparameters tested in the process of model induction.
keras_epochs	Number of epochs used in keras-based methods, if keras methods are used. (e.g. "mlpKerasDropout", "mlpKerasDecay")
search_iters_mxnet	Number of iterations in mxnet-based neural network creation. Default: 5000
cores	Number of cores using in parallel processing.
output_file	Out csv file for the benchmark.
mxnet	Whether to use mxnet. Default: F
gpu	Wheter to use GPU in mxnet and keras processing. Default: F
algorithms	Caret methods that will be checked in benchmark processing. By default the logistic regression is always included.
holdout	Best set of hyperparameters can be selected using: (1) if TURE - using hold-out validation on test set, (2) if FALSE - using 10-fold cross-validation repeated 5 times.
stamp	Character vector or timestamp to make the benchmark unique.
input_fomulas	List of formulas as created by 'ks.miRNAselector' or 'ks.merge_formulas'. Those formulas will be check in benchmark.
keras_threds	This package supports training of keras networks in parallel. Here you can set the number of threads used. (e.g. "mlpKerasDropout", "mlpKerasDecay")

Value

Results of benchmark. Note that benchmark files are also saved in working directory ('wd').

kstawiski/miRNAselector documentation built on Oct. 10, 2020, 9:03 a.m.