Cubist: "subscript out of bounds" or "evaluation nested too deeply"

[frankiethull]

The problem

I'm having trouble with cubist depending which computer I use. I was trying to reprex the second issue and stumbled into the first.

1. converting cubist fits to orbital objects created a "subscript out of bounds" error on my linux laptop, 2) that conversion worked on my work laptop, but there was a downstream issue with orbital_sql/predict that stated "evaluation nested too deeply" which is a windows laptop.

Reproducible example

################################################################
#                       cubist orb 
################################################################
library(Cubist)
#> Loading required package: lattice
library(rules)
#> Loading required package: parsnip
library(orbital)
library(duckdb)
#> Loading required package: DBI

model_df <- modeldata::penguins |> na.omit()

crf_spec <- parsnip::cubist_rules(
  committees = 95, 
  max_rules = 99,
  neighbors = 2) |>
  parsnip::set_engine("Cubist") |>
  parsnip::set_mode("regression")

crf_rec <- recipes::recipe(body_mass_g ~ ., model_df) |>
  recipes::step_dummy(recipes::all_nominal_predictors()) |>
  recipes::step_nzv(recipes::all_predictors()) 


crf_wkflw <- workflows::workflow() |>
  workflows::add_model(crf_spec) |>
  workflows::add_recipe(crf_rec)

crf_fit <- crf_wkflw |>
  workflows::fit(model_df)


crf_orb <- orbital::orbital(crf_fit)
#> Error in `map()`:
#> ℹ In index: 4.
#> Caused by error in `path[[1]]`:
#> ! subscript out of bounds

################################################################
#                    in memory duckdb 
################################################################

penquack_con <- dbConnect(duckdb(dbdir = ":memory:"))
model_table <- dplyr::copy_to(penquack_con, model_df)

# TODO, recreate `Error: evaluation nested too deeply: infinite recursion / options(expressions=)?`
crf_sql <- orbital::orbital_sql(crf_orb, penquack_con)
#> Error: object 'crf_orb' not found

### w/ orbital
tictoc::tic()
# TODO, recreate `Error: evaluation nested too deeply: infinite recursion / options(expressions=)?`
predict(crf_orb, model_table)
#> Error: object 'crf_orb' not found
tictoc::toc()
#> 0.001 sec elapsed

### w/o orbital
tictoc::tic()
predict(crf_fit, model_df)
#> # A tibble: 333 × 1
#>    .pred
#>    <dbl>
#>  1 3682.
#>  2 3532.
#>  3 3345.
#>  4 3554.
#>  5 3691.
#>  6 3540.
#>  7 4126.
#>  8 3419.
#>  9 3759.
#> 10 4144.
#> # ℹ 323 more rows
tictoc::toc()
#> 0.032 sec elapsed

^{Created on 2025-07-30 with reprex v2.1.1}

[EmilHvitfeldt]

Hello @frankiethull 👋

Thanks for reporting, I'll take a look at these issues, I have seen both types of issues for other models/engines so I hope they will be similarly fixed

[EmilHvitfeldt]

I have been able to reproduce with simpler model

library(Cubist)
#> Loading required package: lattice
library(rules)
#> Loading required package: parsnip
library(orbital)

model_df <- modeldata::penguins |> na.omit()

crf_spec <- parsnip::cubist_rules(
  committees = 2, 
  max_rules = 5,
  neighbors = 2) |>
  parsnip::set_engine("Cubist") |>
  parsnip::set_mode("regression")

crf_rec <- recipes::recipe(body_mass_g ~ ., model_df) |>
  recipes::step_dummy(recipes::all_nominal_predictors()) |>
  recipes::step_nzv(recipes::all_predictors()) 


crf_wkflw <- workflows::workflow() |>
  workflows::add_model(crf_spec) |>
  workflows::add_recipe(crf_rec)

crf_fit <- crf_wkflw |>
  workflows::fit(model_df)

crf_orb <- orbital::orbital(crf_fit)
#> Error in `map()`:
#> ℹ In index: 4.
#> Caused by error in `path[[1]]`:
#> ! subscript out of bounds

^{Created on 2025-10-15 with reprex v2.1.1}

[EmilHvitfeldt]

see the following output of summary() of a cubist model.

Model 1:

  Rule 1/1: [199 cases, mean 3674.7, range 2700 to 4775, est err 247.7]

    if
	flipper_length_mm <= 202
    then
	outcome = -320.8 + 405 sex_male + 13.8 flipper_length_mm
	          + 64 bill_depth_mm - 17 island_Dream

Model 2:

  Rule 2/1: [333 cases, mean 4207.1, range 2700 to 6300, est err 271.2]

	outcome = -3358.8 + 37.6 flipper_length_mm + 415 sex_male
	          - 388 island_Dream - 332 island_Torgersen

notice how Rule 2/1 doesn't have a conditional? This is what caused the first error we saw. That is now fixed upstream in tidymodels/tidypredict#128.

The issue with evaluation nested too deeply will be handled and tracked in #105.