evaluateCV: Performance evaluation by cross-validation

In jpvert/kmr4toxicogenetics: Kernel multitask regression for toxicogenetics

Description

Evaluate regression performance of a predictor by cross-validation.

Usage

evaluateCV(mypredictor = c("predictorKMR", "predictorElasticNet",
  "predictorLasso", "predictorRF"), celllines, celllinesKernel, chemicals,
  chemicalsKernel, toxicity, nfolds = 5, nrepeats = 10, seed = 47,
  mc.cores = 1, ...)

Arguments

mypredictor	Character indicating a predictor function. Possible options are KMR (default), ElasticNet, Lasso and RF.
celllines	Matrix of descriptors for ncell cell lines, of dimension ncell x pcell. Only used for ElasticNet, Lasso and RF.
celllinesKernel	Kernel Gram matrix for ncell cell lines, of dimension ncell x ncell. Only used for KMR.
chemicals	Matrix of descriptors for nchem chemicals, of dimension nchem x pchem. Only used for ElasticNet, Lasso and RF.
chemicalsKernel	Kernel Gram matrix for nchem chemicals, of dimension nchem x nchem. Only used for KMR.
toxicity	Matrix of toxicity values for ncell cell lines responding to nchem chemicals, of dimension ncell x nchem.
nfolds	Number of folds for cross-validation. Default is 5.
nrepeats	Number of times the k-fold cross-validation is performed. Default is 1.
seed	A seed number for the random number generator (useful to have the same CV splits).
mc.cores	Number of parallelable CPU cores to use.
...	Other arguments to pass to predictor function.

Value

A list with matrices of cross-validation performance scores. Each score matrix is of dimension nexp x nchem (per CV experiment, per chemical) where nexp=nfolds*nrepeats and corresponds to one of the evaluation criteria:

matrix.ci	Concordance index.
matrix.rho	Pearson correlation.

References

Bernard, E., Jiao, Y., Scornet, E., Stoven, V., Walter, T., and Vert, J.-P. (2017). "Kernel multitask regression for toxicogenetics." bioRxiv-171298.

See Also

predictorKMR, predictorElasticNet, predictorLasso, predictorRF

jpvert/kmr4toxicogenetics documentation built on May 24, 2019, 2:04 a.m.