Merge pull request #344 from tidymodels/tidy-glmnet

better tidy glmnet methods

NAMESPACE

@@ -55,6 +55,10 @@ S3method(print,svm_poly)
 S3method(print,svm_rbf)
 S3method(req_pkgs,model_fit)
 S3method(req_pkgs,model_spec)
+S3method(tidy,"_elnet")
+S3method(tidy,"_fishnet")
+S3method(tidy,"_lognet")
+S3method(tidy,"_multnet")
 S3method(tidy,model_fit)
 S3method(tidy,nullmodel)
 S3method(translate,boost_tree)
@@ -234,6 +238,7 @@ importFrom(stats,.checkMFClasses)
 importFrom(stats,.getXlevels)
 importFrom(stats,as.formula)
 importFrom(stats,binomial)
+importFrom(stats,coef)
 importFrom(stats,delete.response)
 importFrom(stats,model.frame)
 importFrom(stats,model.matrix)

NEWS.md

@@ -1,5 +1,7 @@
 # parsnip (development version)
 
+ * Specific `tidy()` methods for `glmnet` models fit via `parsnip` were created so that the coefficients for the specific fitted `parsnip` model are returned.
+
 # parsnip 0.1.2
 
 ## Breaking Changes

R/aaa.R

@@ -40,7 +40,7 @@ utils::globalVariables(
     "neighbors", ".submodels", "has_submodel", "max_neighbor", "max_penalty",
     "max_terms", "max_tree", "model", "name", "num_terms", "penalty", "trees",
     "sub_neighbors", ".pred_class", "x", "y", "predictor_indicators",
-    "compute_intercept", "remove_intercept")
+    "compute_intercept", "remove_intercept", "estimate", "term")
 )
 
 # nocov end

R/predict.R

@@ -81,12 +81,12 @@
 #' lm_model <-
 #'   linear_reg() %>%
 #'   set_engine("lm") %>%
-#'   fit(mpg ~ ., data = mtcars %>% slice(11:32))
+#'   fit(mpg ~ ., data = mtcars %>% dplyr::slice(11:32))
 #'
 #' pred_cars <-
 #'   mtcars %>%
-#'   slice(1:10) %>%
-#'   select(-mpg)
+#'   dplyr::slice(1:10) %>%
+#'   dplyr::select(-mpg)
 #'
 #' predict(lm_model, pred_cars)
 #'

R/tidy_glmnet.R

@@ -0,0 +1,63 @@
+#' tidy methods for glmnet models
+#'
+#' `tidy()` methods for the various `glmnet` models that return the coefficients
+#' for the specific penalty value used by the `parsnip` model fit.
+#' @param x A fitted `parsnip` model that used the `glmnet` engine.
+#' @param penalty A _single_ numeric value. If none is given, the value specified
+#' in the model specification is used.
+#' @param ... Not used
+#' @return A tibble with columns `term`, `estimate`, and `penalty`. When a
+#' multinomial mode is used, an additional `class` column is included.
+#' @importFrom stats coef
+#' @export
+tidy._elnet <- function(x, penalty = NULL, ...) {
+  tidy_glmnet(x, penalty)
+}
+
+#' @export
+#' @rdname tidy._elnet
+tidy._lognet <- function(x, penalty = NULL, ...) {
+  tidy_glmnet(x, penalty)
+}
+
+#' @export
+#' @rdname tidy._elnet
+tidy._multnet <- function(x, penalty = NULL, ...) {
+  tidy_glmnet(x, penalty)
+}
+
+#' @export
+#' @rdname tidy._elnet
+tidy._fishnet <- function(x, penalty = NULL, ...) {
+  tidy_glmnet(x, penalty)
+}
+
+## -----------------------------------------------------------------------------
+
+get_glmn_coefs <- function(x, penalty = 0.01) {
+  res <- coef(x, s = penalty)
+  res <- as.matrix(res)
+  colnames(res) <- "estimate"
+  rn <- rownames(res)
+  res <- tibble::as_tibble(res) %>% mutate(term = rn, penalty = penalty)
+  res <- dplyr::select(res, term, estimate, penalty)
+  if (is.list(res$estimate)) {
+    res$estimate <- purrr::map(res$estimate, ~ as_tibble(as.matrix(.x), rownames = "term"))
+    res <- tidyr::unnest(res, cols = c(estimate), names_repair = "minimal")
+    names(res) <- c("class", "term", "estimate", "penalty")
+  }
+  res
+}
+
+tidy_glmnet <- function(x, penalty = NULL, ...) {
+  check_installs(x$spec)
+  load_libs(x$spec, quiet = TRUE, attach = TRUE)
+  if (is.null(penalty)) {
+    if (isTRUE(is.numeric(x$spec$args$penalty))){
+      penalty <- x$spec$args$penalty
+    } else {
+      rlang::abort("Please pick a single value of `penalty`.")
+    }
+  }
+  get_glmn_coefs(x$fit, penalty = penalty)
+}

R/zzz.R

@@ -3,6 +3,10 @@
 .onLoad <- function(libname, pkgname) {
   s3_register("broom::tidy", "model_fit")
   s3_register("broom::tidy", "nullmodel")
+  s3_register("broom::tidy", "_elnet")
+  s3_register("broom::tidy", "_lognet")
+  s3_register("broom::tidy", "_multnet")
+  s3_register("broom::tidy", "_fishnet")
 }
 
 

tests/testthat/test_multinom_reg_glmnet.R

@@ -143,8 +143,8 @@ test_that('glmnet probabilities, mulitiple lambda', {
 
   for (i in seq_along(mult_class_res$.pred)) {
     expect_equal(
-      mult_class     %>% slice(i) %>% pull(.pred) %>% purrr::pluck(1) %>% dplyr::select(starts_with(".pred")),
-      mult_class_res %>% slice(i) %>% pull(.pred) %>% purrr::pluck(1) %>% dplyr::select(starts_with(".pred"))
+      mult_class     %>% dplyr::slice(i) %>% pull(.pred) %>% purrr::pluck(1) %>% dplyr::select(starts_with(".pred")),
+      mult_class_res %>% dplyr::slice(i) %>% pull(.pred) %>% purrr::pluck(1) %>% dplyr::select(starts_with(".pred"))
     )
   }
 

tests/testthat/test_tidy_glmnet.R

@@ -0,0 +1,58 @@
+context("tidy glmnet models")
+
+test_that('linear regression', {
+  skip_if_not_installed("glmnet")
+
+  ps_mod <-
+    linear_reg(penalty = .1) %>%
+    set_engine("glmnet") %>%
+    fit(mpg ~ ., data = mtcars)
+
+  ps_coefs <- tidy(ps_mod)
+  gn_coefs <- as.matrix(coef(ps_mod$fit, s = .1))
+  for(i in ps_coefs$term) {
+    expect_equal(ps_coefs$estimate[ps_coefs$term == i], gn_coefs[i,1])
+  }
+})
+
+test_that('logistic regression', {
+  skip_if_not_installed("glmnet")
+
+  data(two_class_dat, package = "modeldata")
+
+  ps_mod <-
+    logistic_reg(penalty = .1) %>%
+    set_engine("glmnet") %>%
+    fit(Class ~ ., data = two_class_dat)
+
+  ps_coefs <- tidy(ps_mod)
+  gn_coefs <- as.matrix(coef(ps_mod$fit, s = .1))
+  for(i in ps_coefs$term) {
+    expect_equal(ps_coefs$estimate[ps_coefs$term == i], gn_coefs[i,1])
+  }
+})
+
+test_that('multinomial regression', {
+  skip_if_not_installed("glmnet")
+
+  data(penguins, package = "modeldata")
+
+  ps_mod <-
+    multinom_reg(penalty = .01) %>%
+    set_engine("glmnet") %>%
+    fit(species ~ ., data = penguins)
+
+  ps_coefs <- tidy(ps_mod)
+  gn_coefs <- coef(ps_mod$fit, s = .01)
+  gn_coefs <- purrr::map(gn_coefs, as.matrix)
+  for(i in unique(ps_coefs$term)) {
+    for(j in unique(ps_coefs$class)) {
+      expect_equal(
+        ps_coefs$estimate[ps_coefs$term == i & ps_coefs$class == j],
+        gn_coefs[[j]][i,1]
+        )
+    }
+  }
+})
+
+