RandomForest Not Working with Certain Datasets

[washcycle]

Using a RandomForest causes an error from the purr package with certain datasets.

It was on one of the RStudio Conference 2018 "Big Data in R" EC2 machines.

library(randomForest)
model <- randomForest(mpg ~ ., data = mtcars)
tidypredict_sql(model, con) # any connection I tried caused the same error

ERROR
Error: .xis empty, and no.init supplied
which comes from the purr package.

[washcycle]

I debugged it down to right here so far...

parsemodel.R

   all_paths <- model_frame %>%
    filter(.data$status == -1) %>%
    pull(.data$rowid) %>%
    map(~get_path(.x, model_frame)) %>%
    bind_rows()

all_paths comes back empty

[washcycle]

When fitting a random forest for regression the status variables doesn't contain 1. It contains -1 and -3 that might be the issue. Can use randomForest::getTree(...) on the model for a regression and classification case to see what I am talking about.

[washcycle]

I think I figured out the issue. The randomForest::getTree has a different criteria for regression problems.

For numerical predictors, data with values of the variable less than or equal to the splitting point go to the left daughter node.

[edgararuiz-zz]

Fixed with #19 - Thanks @washcycle!

[github-actions]

This issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex: https://reprex.tidyverse.org) and link to this issue.