Activation checks

NEWS.md

@@ -1,5 +1,7 @@
 # parsnip (development version)
 
+* parsnip now lets the engines for [mlp()] check for acceptable values of the activation function (#1019)
+
 * Tightened logic for outcome checking. This resolves issues—some errors and some silent failures—when atomic outcome variables have an attribute (#1060, #1061).
 
 * `rpart_train()` has been deprecated in favor of using `decision_tree()` with the `"rpart"` engine or `rpart::rpart()` directly (#1044).

R/mlp.R

@@ -21,8 +21,8 @@
 #' @param activation A single character string denoting the type of relationship
 #'  between the original predictors and the hidden unit layer. The activation
 #'  function between the hidden and output layers is automatically set to either
-#'  "linear" or "softmax" depending on the type of outcome. Possible values are:
-#'  "linear", "softmax", "relu", and "elu"
+#'  "linear" or "softmax" depending on the type of outcome. Possible values
+#'  depend on the engine being used.
 #'
 #' @templateVar modeltype mlp
 #' @template spec-details
@@ -142,24 +142,6 @@ check_args.mlp <- function(object) {
     if (args$dropout > 0 & args$penalty > 0)
       rlang::abort("Both weight decay and dropout should not be specified.")
 
-
-  if (object$engine == "brulee") {
-    act_funs <- c("linear", "relu", "elu", "tanh")
-  } else if (object$engine == "keras") {
-    act_funs <- c("linear", "softmax", "relu", "elu")
-  } else if (object$engine == "h2o") {
-    act_funs <- c("relu", "tanh")
-  }
-
-  if (is.character(args$activation)) {
-    if (!any(args$activation %in% c(act_funs))) {
-      rlang::abort(
-        glue::glue("`activation` should be one of: ",
-                   glue::glue_collapse(glue::glue("'{act_funs}'"), sep = ", "))
-      )
-    }
-  }
-
   invisible(object)
 }
 
@@ -210,6 +192,9 @@ keras_mlp <-
            seeds = sample.int(10^5, size = 3),
            ...) {
 
+    act_funs <- c("linear", "softmax", "relu", "elu")
+    rlang::arg_match(activation, act_funs,)
+
     if (penalty > 0 & dropout > 0) {
       rlang::abort("Please use either dropoput or weight decay.", call. = FALSE)
     }

tests/testthat/test_mlp.R

@@ -23,3 +23,28 @@ test_that("nnet_softmax", {
   expect_equal(res$b, 1 - res$a)
 })
 
+test_that("more activations for brulee", {
+  skip_if_not_installed("brulee", minimum_version = "0.3.0")
+  skip_on_cran()
+
+  data(ames, package = "modeldata")
+
+  ames$Sale_Price <- log10(ames$Sale_Price)
+
+  set.seed(122)
+  in_train <- sample(1:nrow(ames), 2000)
+  ames_train <- ames[ in_train,]
+  ames_test  <- ames[-in_train,]
+
+  set.seed(1)
+  fit <-
+    try(
+      mlp(penalty = 0.10, activation = "softplus") %>%
+        set_mode("regression") %>%
+        set_engine("brulee") %>%
+        fit_xy(x = as.matrix(ames_train[, c("Longitude", "Latitude")]),
+               y = ames_train$Sale_Price),
+      silent = TRUE)
+  expect_true(inherits(fit$fit, "brulee_mlp"))
+})
+