uniLasso (version 2.11)

simulate_Gaussian: simulate Gaussian data

Description

A simulator that builds a training and test set with particular characteristics, as used in our "uniLasso" paper.

Usage

simulate_Gaussian(
  ntrain = 300,
  ntest = 3000,
  p = 1000,
  snr = 1,
  rho = 0.8,
  sparsity = 0.1,
  homecourt = FALSE
)

Value

a list with components "x", "y", "xtest", "ytest", "mutest", and "sigma", where "mutest" is the true test mean, and "ytest <- mutest + rnorm(ntest)*sigma."

Arguments

ntrain

number of training examples.

ntest

number of test examples.

p

number of features.

snr

desired SNR (signal-to-noise ratio).

rho

for homecourt=TRUE 'rho' controls the autocorrelation between variables. Variables k units apart have correlation rho^k.

sparsity

fraction of variables with nonzero coefficients.

homecourt

logical; if TRUE then correlated features, with a special boost for large coefficients, mimicking the uniLasso two-stage algorithm.

Examples

dat = simulate_Gaussian(300,3000,p=500,snr=1.2)
fit = cv.uniLasso(dat$x, dat$y)
mse = mean( (predict(fit, dat$xtest)- dat$mutest)^2)