R cv.glmnet error: "need at least two non-NA values to interpolate"

Question

I am getting an error that I can't trace when running the cv.glmnet function. Research only hinted me to the approx() function that throws the error when searching for lambda but I am not familiar enough with glmnet to figure out how it connects.

The error is: "need at least two non-NA values to interpolate"

While the solution seems obvious, I am working with simulated data, so definitely no NAs in there. I double-checked multiple times. I also dropped columns with constant variables in one attempt but didn't help either.

Reproducible example is a bit messy due to the complexity of the data (and also not minimal since the error doesn't occur in small example data) but please see below:

pacman::p_load(glmnet,RCurl)

url <- getURL('https://raw.githubusercontent.com/oliverkostorz/MC-Simulation-Ensemble-Methods-Heterogeneous-Treatment-Effects/master/example.csv')
data <- read.csv(text = url)

x <- data[,3:ncol(data)]
y <- data[,2]

EN_fit <- cv.glmnet(x, y, type.measure = 'mse', alpha = .5)

Hope someone can help!

Best, Oliver

Answer 1

9.352584e+229 is causing issues. It's quite a big number.

big_index <- y > 1e120
EN_fit <- cv.glmnet(as.matrix(x[!big_index, ]), y[!big_index], 
    type.measure = 'mse', alpha = .5)
EN_fit
# 
# Call:  cv.glmnet(x = as.matrix(x[!big_index, ]), y = y[!big_index],
#     type.measure = "mse", alpha = 0.5) 
# 
# Measure: Mean-Squared Error 
# 
#         Lambda Index    Measure  SE Nonzero
# min 3.864e+113    84 8.533e+229 Inf      10
# 1se 1.836e+115     1 8.644e+229 Inf       0