heatcv {ePCR}

Plot a heatmap of the prediction performance statistic as a function of lambda and alpha combinations

Description

This function plots a heatmap of cross-validation results by varying the penalization/regularization parameter (lambda, x-axis), together with the corresponding L1/L2 norm parameter alpha (i.e. LASSO, elastic net, ridge regression). The optimal spot in the parameter grid gives insight into the behavior of the regularization in respect to the norms, but note that the lambda-parameter on x-axis is not constant given a conditional alpha-parameter; rather it is a suitable vector chosen by the glmnet-package.

Usage

heatcv(
  psp,
  bias = 0.1,
  by.rownames = 1,
  by.colnames = 1,
  paletcol = c("cyan", "blue", "black", "red", "orange"),
  paletncol = 1000,
  xlab = "Alpha-dependent log-Lambda",
  ylab = "Alpha",
  main = "",
  plot.opt = TRUE,
  plot.1sd = FALSE,
  ...
)

Arguments

psp

An S4-class PSP-object to plot, as built using the ePCR-package

bias

Bias in color palette (skews it to favor distinguishing high values better by default)

by.rownames

Show every n:th row name (helps for dense axis labels)

by.colnames

Show every n:th column name (helps for dense axis labels)

paletcol

Names for colours to include in the heatmap palette

paletncol

Number of colours on the color key

xlab

Label for the x-axis (typically log-lambda penalization parameter)

ylab

Label for the y-axis (typically alpha-value indicating LASSO, elastic net or ridge regression)

main

Main label on top of the heatmap

plot.opt

Should the best (highest) performance statistic be indicated as a large dot on the heatmap

plot.1sd

Should boundaries of the optimal performance statistic area be outlined as within 1 standard deviation of the optimal spot (note: experimental). This attempts to mimic the 1sd-optimum suggested in the glmnet-package for cross-validation for a constant alpha parameter but for 2 dimensions.

...

additional parameters passed on to the hmap-function of hamlet-package

Note

The heatmap plotting is compatible with the default plot-region in a R graphic canvas. The function hmap from the same author's hmap-package can be highly customized to fit more specific needs.

Author(s)

Teemu Daniel Laajala [e-mail]

Examples

data(ePCRmodels)
par(mfrow=c(1,3))
heatcv(DREAM@PSPs[[1]], main=DREAM@PSPs[[1]]@description, by.rownames=10, by.colnames=10)
heatcv(DREAM@PSPs[[2]], main=DREAM@PSPs[[2]]@description, by.rownames=10, by.colnames=10)
heatcv(DREAM@PSPs[[3]], main=DREAM@PSPs[[3]]@description, by.rownames=10, by.colnames=10)