How to make all interactions before using glmnet

Question

I have an x-matrix of 8 columns. I want to run glmnet to do a lasso regression. I know I need to call:

glmnet(x, y, family = "binomial", ...).

However, how do I get x to consider all one way interactions as well? Do I have to manually remake the data frame: if so, is there an easier way? I suppose I was hoping to do something using an R formula.

score 37 · Accepted Answer · answered Dec 20 '14 at 20:14

37

Yes, there is a convenient way for that. Two steps in it are important.

library(glmnet)
# Sample data
data <- data.frame(matrix(rnorm(9 * 10), ncol = 9))
names(data) <- c(paste0("x", 1:8), "y")
# First step: using .*. for all interactions
f <- as.formula(y ~ .*.)
y <- data$y
# Second step: using model.matrix to take advantage of f
x <- model.matrix(f, data)[, -1]
glmnet(x, y)

answered Dec 20 '14 at 20:14

Julius Vainora

47,421
9
90
102

1

[, -1] is to remove an 'intercept' column that is auto created using model.matrix in this example. – theforestecologist Jul 24 '15 at 16:46
is it possible to do this with caret? When I feed the model matrix into caret train with same settings it doesn't run the interaction variables – KillerSnail Feb 22 '16 at 16:11
@KillerSnail, try `f <- as.formula( ~ .*.)` and `x <- model.matrix(f, TrainData)[, -1]` and then take `train(x = x, ...)`. – Julius Vainora Feb 25 '16 at 18:11
@RobertKubrick even though glmnet scales them before estimating? – suckrates Dec 10 '18 at 15:40
Is there an easy way to multiply column 1 by columns 2:10? – Dave Jun 20 '19 at 00:27
@Dave, `x[, 1] * x[, 2:10]` does what you said; if you have something else in mind you should post a question. – Julius Vainora Jun 20 '19 at 00:30
1

I think this doesn't include quadratic terms? I.e. self-interactions. – jiggunjer Jul 16 '20 at 09:34

score 3 · Answer 2 · answered Nov 22 '16 at 18:59

3

f <- as.formula( ~ .^2) should also work for including main effects and all pairwise interactions

answered Nov 22 '16 at 18:59

bgreenwell

378
2
8

How to make all interactions before using glmnet

2 Answers2

Linked