Brute force model search with a min and max number of features.

Ranks models based on K-fold CV Q2

Usage

model.subset(
  data,
  out.col = dim(data)[2],
  min = 2,
  max = floor(dim(data)[1]/5),
  folds = nrow(data),
  iterations = 1,
  cutoff = 0.85,
  cross.terms = F
)

Arguments

data

data frame with output column

out.col

number of output column

min

minimum # of features (default = 2)

max

max # of features (defaults = # of observations / 5)

folds

defaults to nrow(data)

iterations

defaults to 1 (LOOCV)

cutoff

search for Q2 above 0.85 (if there isn't will look for lower)

cross.terms

if TRUE includes feature interactions (explosive - try avoiding)

Value

table with 10 best models (at max)