example

ModelOriented · Jan 9, 2018 · 33fd052 · 33fd052
1 parent a591c7e
commit 33fd052
Show file tree

Hide file tree

Showing 2 changed files with 22 additions and 4 deletions.
diff --git a/R/explain.R b/R/explain.R
@@ -66,14 +66,17 @@ prepare_forestplot <- function(model) {
 #'
 #' @return plot (ggplot2 or base)
 #'
+
 plot_regression <- function(plot_type, fitted_model, explained_instance, scale = NULL) {
   if(plot_type == "forestplot") {
     prepare_forestplot(fitted_model)
   } else {
     if(scale == "probability") {
       plot(breakDown::broken(fitted_model, explained_instance, baseline = "intercept"),
            trans = function(x) exp(x)/(1 + exp(x))) +
-        scale_y_continuous(limits = c(0, 1), name = "probability", expand = c(0, 0))
+      ggplot2::scale_y_continuous(limits = c(0, 1), 
+                                  name = "probability", 
+                                  expand = c(0, 0))
 
     } else {
       plot(breakDown::broken(fitted_model, explained_instance, baseline = "intercept"))

diff --git a/vignettes/HR.Rmd b/vignettes/HR.Rmd
@@ -21,7 +21,7 @@ trees <- randomForest(left~., data = HR_data, ntree=1000)
 similar <- sample_locally(data = HR_data,
                           explained_instance = HR_data[2,],
                           explained_var = "left", 
-                          size = 200)
+                          size = 2000)
 head(similar$data)
 
 similar <- add_predictions(HR_Data, similar, black_box_model = trees)
@@ -32,8 +32,23 @@ trained <- fit_explanation(live_object = similar,
                            selection = FALSE)
 # trained
 plot_explanation(trained, "waterfallplot", 
-                 explained_instance = HR_data[1,],
+                 explained_instance = HR_data[2,],
                  scale = "probability")
 plot_explanation(trained, "forestplot", 
-                 explained_instance = HR_data[1,])
+                 explained_instance = HR_data[2,])
+
+HR_data$left <- as.numeric(as.character(HR_data$left))
+trees <- randomForest(left~., data = HR_data, ntree=1000)
+
+similar2 <- sample_locally(data = HR_data,
+                           explained_instance = HR_data[2,],
+                           explained_var = "left", 
+                           size = 2000)
+similar2 <- add_predictions(HR_Data, similar2, black_box_model = trees)
+glimpse(similar2$data)
+trained2 <- fit_explanation(live_object = similar2,
+                            white_box = "regr.lm",
+                            selection = F)
+plot_explanation(trained2, "forestplot", 
+                 explained_instance = HR_data[2,])
 ```