Repository navigation
subset_covariates does not successfully reorder covariates to match training structure #428
Description
Error is here: https://github.com/tlverse/sl3/blob/master/R/Lrnr_base.R#L75-L78
This can cause errors in prediction when covariates are reordered relative to training.
Example:
data(mtcars)
lrnr <- Lrnr_glmnet$new()
task1 <- make_sl3_Task(mtcars, outcome="mpg", covariates = c("hp","disp","cyl"))
task2 <- make_sl3_Task(mtcars, outcome="mpg", covariates = c("disp","cyl","hp"))
fit <- lrnr$train(task1)
fit$predict(task1)
fit$predict(task2)
> fit$predict(task1)
[1] 22.17332 22.17332 25.84009 20.33786 15.04299 21.02797 14.03433 25.56196 25.19695 21.84366 21.84366 16.54794 16.54794 16.54794 12.51304 12.59369 12.75214 26.77790 27.03582
[20] 26.93466 25.55582 16.18985 16.45206 14.22162 14.29382 26.77229 25.63854 25.79350 13.92912 21.51765 13.84251 25.36606
> fit$predict(task2)
[1] -163.87883 -163.87883 -99.88659 -283.82522 -409.91359 -243.34146 -411.22464 -146.67257 -140.06937 -173.42427 -173.42427 -306.95126 -306.95126 -306.95126 -547.55706
[16] -533.05704 -508.85912 -63.51938 -59.58534 -54.19868 -114.77121 -358.03977 -340.90457 -398.98521 -458.87130 -63.88656 -114.90363 -84.47231 -400.56501 -146.73708
[31] -340.69764 -116.09751
Activity