feature request: add way to remove intercept in linear_reg() lm

[jospueyo]

Feature: step_rm_intercept()

In situations when a linear model without intercept is needed, recipes does not provide a straightforward method. If you try the conventional way:

library(tidymodels)

rec <- recipe(mpg ~ cyl -1, data = mtcars)
#> Error in `recipe()`:
#> ! `-` is not allowed in a recipe formula. Use `step_rm()` instead.
#> Backtrace:
#>     ▆
#>  1. ├─recipes::recipe(mpg ~ cyl - 1, data = mtcars)
#>  2. └─recipes:::recipe.formula(mpg ~ cyl - 1, data = mtcars)
#>  3.   └─rlang::abort("`-` is not allowed in a recipe formula. Use `step_rm()` instead.")

I found a workaround in stackoverflow, where, by the way, @juliasilge already suggested to include this feature.

However, I do not know how to integrate this workaround when multiple models are set using workflow_set because the add_model step, suggested in the above answer, is not used:

library(tidymodels)

rec1 <- recipe(mpg ~ cyl + hp, data = mtcars)
rec2 <- rec1 |>
  step_rm(hp)

workflow_set(list(rec1, rec2), list(lm = linear_reg()))
#> # A workflow set/tibble: 2 × 4
#>   wflow_id    info             option    result    
#>   <chr>       <list>           <list>    <list>    
#> 1 recipe_1_lm <tibble [1 × 4]> <opts[0]> <list [0]>
#> 2 recipe_2_lm <tibble [1 × 4]> <opts[0]> <list [0]>

^{Created on 2023-05-18 with reprex v2.0.2}

Thanks for the package!

[EmilHvitfeldt]

Hello @jospueyo 👋

Thank you for your feature request! The problem arrises in parsnip, as it receives the data that comes out of recipes(). At that stage a recipe is generated using the outcome and predictors.

What would be an better option in my opinion is to give us a way to remove the intercept in lm() without doing so through a formula

[simonpcouch]

Thanks for the issue, @jospueyo! The solution here is to use workflows' model formula. From workflows' docs:

formula: An optional formula override to specify the terms of the model. Typically, the terms are extracted from the formula or recipe preprocessing methods. However, some models (like survival and bayesian models) use the formula not to preprocess, but to specify the structure of the model. In those cases, a formula specifying the model structure must be passed unchanged into the model call itself. This argument is used for those purposes.

So, mpg ~ cyl + hp specifies the terms, mpg ~ cyl + hp - 1 specifies the model structure which lm() knows how to interpret.

For example:

library(parsnip)
library(workflows)

# via base:
lm(mpg ~ cyl + hp - 1, mtcars)
#> 
#> Call:
#> lm(formula = mpg ~ cyl + hp - 1, data = mtcars)
#> 
#> Coefficients:
#>     cyl       hp  
#>  5.4036  -0.1075

# via parsnip
linear_reg() %>% fit(mpg ~ cyl + hp - 1, mtcars)
#> parsnip model object
#> 
#> 
#> Call:
#> stats::lm(formula = mpg ~ cyl + hp - 1, data = data)
#> 
#> Coefficients:
#>     cyl       hp  
#>  5.4036  -0.1075

# via workflows
workflow() %>%
  add_formula(mpg ~ cyl) %>%
  add_model(linear_reg(), formula = mpg ~ cyl + hp - 1) %>%
  fit(mtcars)
#> Error in eval(predvars, data, env): object 'hp' not found

^{Created on 2023-11-02 with reprex v2.0.2}

[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.