...

book/chapters/chapter15/predsets_valid_inttune.qmd

@@ -11,7 +11,7 @@ aliases:
 
 ## Predict Sets and Training Error Estimation {#sec-predict-sets}
 
-In the @sec-performance chapter we have already studied in detail how to train, predict and evaluate many different learners. Evaluating a fully trained model usually requires making predictions on unseen test observations. When we predict directly with a trained learner, we can explicitly control which observations are used:
+In @sec-performance we have already studied in detail how to train, predict and evaluate many different learners. Evaluating a fully trained model usually requires making predictions on unseen test observations. When we predict directly with a trained learner, we can explicitly control which observations are used:
 
 ```{r}
 tsk_sonar = tsk("sonar")
@@ -255,10 +255,9 @@ data = lrn_xgb$model$evaluation_log
 ggplot(data, aes(x = iter, y = test_logloss)) +
   geom_line() +
   geom_point(data = data.table(x = lrn_xgb$internal_tuned_values$nrounds,
-    y = lrn_xgb$internal_valid_scores$logloss), aes(x = x, y = y, color = "red")) +
+    y = lrn_xgb$internal_valid_scores$logloss), aes(x = x, y = y, color = "red"), show.legend = FALSE) +
   labs(
-    x = "Iteration", y = "Validation Logloss",
-    color = NULL
+    x = "Iteration", y = "Validation Logloss"
   )
 ```