Merge pull request #310 from SchlossLab/feat-imp

Show example feature importance plot

NEWS.md

@@ -1,7 +1,8 @@
 # mikropml development version
 
+- Created an example showing how to plot feature importances in the `parallel` vignette (#310, @kelly-sovacool).
 - You can now use `parRF`, a parallel implementation of the `rf` method, with
-  the same default hyperparameters as `rf` set automatically.
+  the same default hyperparameters as `rf` set automatically (#306, @kelly-sovacool).
 
 # mikropml 1.4.0
 

R/feature_importance.R

@@ -3,7 +3,6 @@
 #' Calculates feature importance using a trained model and test data. Requires
 #' the `future.apply` package.
 #'
-#' @param train_data Training data: dataframe of outcome and features.
 #' @inheritParams run_ml
 #' @inheritParams calc_perf_metrics
 #' @inheritParams group_correlated_features
@@ -121,15 +120,15 @@
 #' @author Begüm Topçuoğlu, \email{topcuoglu.begum@@gmail.com}
 #' @author Zena Lapp, \email{zenalapp@@umich.edu}
 #' @author Kelly Sovacool, \email{sovacool@@umich.edu}
-get_feature_importance <- function(trained_model, train_data, test_data,
+get_feature_importance <- function(trained_model, test_data,
                                    outcome_colname, perf_metric_function,
                                    perf_metric_name, class_probs, method,
                                    seed = NA, corr_thresh = 1, groups = NULL,
                                    nperms = 100, corr_method = "spearman") {
   abort_packages_not_installed("future.apply")
 
   # get outcome and features
-  split_dat <- split_outcome_features(train_data, outcome_colname)
+  split_dat <- split_outcome_features(test_data, outcome_colname)
   outcome <- split_dat$outcome
   features <- split_dat$features
 

R/run_ml.R

@@ -253,7 +253,6 @@ run_ml <-
       message("Finding feature importance...")
       feature_importance_tbl <- get_feature_importance(
         trained_model_caret,
-        train_data,
         test_data,
         outcome_colname,
         perf_metric_function,

R/train_model.R

@@ -58,7 +58,8 @@ train_model <- function(train_data,
           ...
         )
       } else {
-        features_train <- train_data %>% dplyr::select(-outcome_colname)
+        features_train <- train_data %>%
+          dplyr::select(-dplyr::all_of(outcome_colname))
         outcomes_train <-
           train_data %>% dplyr::pull(outcome_colname)
         if (is.character(outcomes_train)) {

README.md

@@ -54,10 +54,10 @@ mamba install -c conda-forge r-mikropml
 
 ### Dependencies
 
--   Imports: caret, dplyr, e1071, glmnet, kernlab, MLmetrics,
-    randomForest, rlang, rpart, stats, utils, xgboost
--   Suggests: doFuture, foreach, future, future.apply, ggplot2, knitr,
-    progress, progressr, purrr, rmarkdown, testthat, tidyr
+- Imports: caret, dplyr, e1071, glmnet, kernlab, MLmetrics,
+  randomForest, rlang, rpart, stats, utils, xgboost
+- Suggests: doFuture, foreach, future, future.apply, ggplot2, knitr,
+  progress, progressr, purrr, rmarkdown, testthat, tidyr
 
 ## Usage