Divert linear regressions with poisson family to poisson_reg() (#1219)

* Changes for #956

* update news

* point to parsnip PR instead of tune issue

* move poisson checks to check_args

* the snapshots should be created without glmnet installed

* GHA for tests has glmnet installed (◔_◔)

---------

Co-authored-by: Hannah Frick <[email protected]>

Author: topepo

NEWS.md

@@ -23,6 +23,8 @@
 
 * Aligned `null_model()` with other model types; the model type now has an engine argument that defaults to `"parsnip"` and is checked with the same machinery that checks other model types in the package (#1083).
 
+* If linear regression is requested with a Poisson family, an error will occur and refer the user to `poisson_reg()` (#1219).
+
 * The deprecated function `rpart_train()` was removed after its deprecation period (#1044).
 
 ## Bug Fixes

R/linear_reg.R

@@ -73,6 +73,7 @@ translate.linear_reg <- function(x, engine = x$engine, ...) {
     # evaluated value for the parameter.
     x$args$penalty <- rlang::eval_tidy(x$args$penalty)
   }
+
   x
 }
 
@@ -113,5 +114,26 @@ check_args.linear_reg <- function(object, call = rlang::caller_env()) {
   check_number_decimal(args$mixture, min = 0, max = 1, allow_null = TRUE, call = call, arg = "mixture")
   check_number_decimal(args$penalty, min = 0, allow_null = TRUE, call = call, arg = "penalty")
 
+  # ------------------------------------------------------------------------------
+  # We want to avoid folks passing in a poisson family instead of using
+  # poisson_reg(). It's hard to detect this.
+
+  is_fam <- names(object$eng_args) == "family"
+  if (any(is_fam)) {
+    eng_args <- rlang::eval_tidy(object$eng_args[[which(is_fam)]])
+    if (is.function(eng_args)) {
+      eng_args <- try(eng_args(), silent = TRUE)
+    }
+    if (inherits(eng_args, "family")) {
+      eng_args <- eng_args$family
+    }
+    if (eng_args == "poisson") {
+      cli::cli_abort(
+        "A Poisson family was requested for {.fn linear_reg}. Please use
+        {.fn poisson_reg} and the engines in the {.pkg poissonreg} package.",
+        call = rlang::call2("linear_reg"))
+    }
+  }
+
   invisible(object)
 }

tests/testthat/_snaps/linear_reg.md

@@ -139,3 +139,39 @@
       Error in `fit()`:
       ! `penalty` must be a number larger than or equal to 0 or `NULL`, not the number -1.
 
+# prevent using a Poisson family
+
+    Code
+      linear_reg(penalty = 1) %>% set_engine("glmnet", family = poisson) %>% fit(mpg ~
+        ., data = mtcars)
+    Condition
+      Error in `fit()`:
+      ! Please install the glmnet package to use this engine.
+
+---
+
+    Code
+      linear_reg(penalty = 1) %>% set_engine("glmnet", family = stats::poisson) %>%
+        fit(mpg ~ ., data = mtcars)
+    Condition
+      Error in `fit()`:
+      ! Please install the glmnet package to use this engine.
+
+---
+
+    Code
+      linear_reg(penalty = 1) %>% set_engine("glmnet", family = stats::poisson()) %>%
+        fit(mpg ~ ., data = mtcars)
+    Condition
+      Error in `fit()`:
+      ! Please install the glmnet package to use this engine.
+
+---
+
+    Code
+      linear_reg(penalty = 1) %>% set_engine("glmnet", family = "poisson") %>% fit(
+        mpg ~ ., data = mtcars)
+    Condition
+      Error in `fit()`:
+      ! Please install the glmnet package to use this engine.
+

tests/testthat/test-linear_reg.R

@@ -358,3 +358,32 @@ test_that("check_args() works", {
     }
   )
 })
+
+
+test_that("prevent using a Poisson family", {
+  skip_if(rlang::is_installed("glmnet"))
+  expect_snapshot(
+    linear_reg(penalty = 1) %>%
+      set_engine("glmnet", family = poisson) %>%
+      fit(mpg ~ ., data = mtcars),
+    error = TRUE
+  )
+  expect_snapshot(
+    linear_reg(penalty = 1) %>%
+      set_engine("glmnet", family = stats::poisson) %>%
+      fit(mpg ~ ., data = mtcars),
+    error = TRUE
+  )
+  expect_snapshot(
+    linear_reg(penalty = 1) %>%
+      set_engine("glmnet", family = stats::poisson()) %>%
+      fit(mpg ~ ., data = mtcars),
+    error = TRUE
+  )
+  expect_snapshot(
+    linear_reg(penalty = 1) %>%
+      set_engine("glmnet", family = "poisson") %>%
+      fit(mpg ~ ., data = mtcars),
+    error = TRUE
+  )
+})