`lm()` doesn't like it when you can't invert the QR matrix

[EmilHvitfeldt]

Simple example here. Gets annoyed during parse_model(). I wonder if we can rewrite this model in orbital to avoid parsing.

library(tidymodels)
library(orbital)

rec_spec <- recipe(Sale_Price ~ ., data = ames) |>
  step_nzv(all_numeric_predictors()) |>
  step_normalize(all_numeric_predictors()) |>
  step_dummy(all_nominal_predictors()) |>
  step_nzv(all_predictors())

mod_spec <- linear_reg()

wf_spec <- workflow(rec_spec, mod_spec)

wf_fit <- fit(wf_spec, ames)

orb <- orbital(wf_fit)
#> Error in qr.solve(qr.R(model$qr)): singular matrix 'a' in solve

Can get working model with:

library(tidymodels)
library(orbital)

rec_spec <- recipe(Sale_Price ~ ., data = ames) |>
  step_normalize(all_numeric_predictors()) |>
  step_dummy(all_nominal_predictors()) |>
  step_zv(all_predictors()) |>
  step_corr(all_predictors())

mod_spec <- linear_reg()

wf_spec <- workflow(rec_spec, mod_spec)

wf_fit <- fit(wf_spec, ames)

orb <- orbital(wf_fit)

[EmilHvitfeldt]

Comes from linear combination, and could be fixed with:

Maybe throw more informative error somewhere

rec_spec <- recipe(Sale_Price ~ ., data = ames) |>
  step_normalize(all_numeric_predictors()) |>
  step_dummy(all_nominal_predictors()) |>
  step_lincomb(all_predictors())

[EmilHvitfeldt]

could throw a better worded message in tidypredict to tell user what went wrong