Projpred varsel specifying search_terms

Modeling
1

Could somebody show me how to specify the search_terms list in the varsel() function?

Here is a minimal example, model first …

library(rstanarm)

n ← 30
d ← 5
x ← matrix(rnorm(nd), nrow=n)
y ← x[,1] + 0.5rnorm(n)
data ← data.frame(x,y)
fit ← stan_glm(y ~ X1 + X2 + X3 + X4 + X5,
gaussian(),
data=data)

Then varsel on all terms in the model works OK.

library(projpred)
vs1 ← varsel(fit)
vs1

But how to select a subset of terms?

vs2 ← varsel(fit, search_terms=c(“X1”,“X2”,“X3”), method=“forward”)

vs3 ← varsel(fit, search_terms=list(“X1”,“X2”,“X3”), method=“forward”)

tt ← terms(fit)
term_list ← as.list(attr(tt, “variables”))
vs4 ← varsel(fit, search_terms=tt, method=“forward”)

All of these show the same error message
Error in sub[“kl”, i] : incorrect number of dimensions

What am I doing wrong?

2

Sorry for the delay. We’re lacking a proper documentation and vignette for search_terms. There’s some discussion at Advice on using search_terms in projpred - #4 by AlejandroCatalina

3

I have collected this and similar questions here. Perhaps the connection to related threads helps until this is resolved in the projpred package (e.g., by an explanation in the documentation).

4

Also sorry for delay for reply from me!

I could not find much in the forum searching for “varsel”. Searching on projpred shows a lot more about using varsel.

I have been looking at this in more detail– I will post something on the forum in the next day or so.

image002.jpg

5

@ACB, you need to include the intercept ("1") in search_terms. If you do this and try again with projpred’s most recent CRAN (or GitHub, shouldn’t matter) version, do you still get the error sub["kl", i] : incorrect number of dimensions?

1 Like