Added support for passing early_stopping_rounds to initializer for Scikit-Learn API

Author: ankane

lib/xgboost/classifier.rb

@@ -18,7 +18,7 @@ def fit(x, y, eval_set: nil, early_stopping_rounds: nil, verbose: true)
 
       @booster = XGBoost.train(params, dtrain,
         num_boost_round: @n_estimators,
-        early_stopping_rounds: early_stopping_rounds,
+        early_stopping_rounds: early_stopping_rounds || @early_stopping_rounds,
         verbose_eval: verbose,
         evals: evals
       )

lib/xgboost/model.rb

@@ -2,10 +2,11 @@ module XGBoost
   class Model
     attr_reader :booster
 
-    def initialize(n_estimators: 100, importance_type: "gain", **options)
+    def initialize(n_estimators: 100, importance_type: "gain", early_stopping_rounds: nil, **options)
       @params = options
       @n_estimators = n_estimators
       @importance_type = importance_type
+      @early_stopping_rounds = early_stopping_rounds
     end
 
     def predict(data)

lib/xgboost/regressor.rb

@@ -10,7 +10,7 @@ def fit(x, y, eval_set: nil, early_stopping_rounds: nil, verbose: true)
 
       @booster = XGBoost.train(@params, dtrain,
         num_boost_round: @n_estimators,
-        early_stopping_rounds: early_stopping_rounds,
+        early_stopping_rounds: early_stopping_rounds || @early_stopping_rounds,
         verbose_eval: verbose,
         evals: evals
       )

test/classifier_test.rb

@@ -50,8 +50,8 @@ def test_multiclass
   def test_early_stopping
     x_train, y_train, x_test, y_test = multiclass_data
 
-    model = XGBoost::Classifier.new
-    model.fit(x_train, y_train, early_stopping_rounds: 5, eval_set: [[x_test, y_test]], verbose: false)
+    model = XGBoost::Classifier.new(early_stopping_rounds: 5)
+    model.fit(x_train, y_train, eval_set: [[x_test, y_test]], verbose: false)
     assert_equal 18, model.booster.best_iteration
   end
 

test/regressor_test.rb

@@ -23,8 +23,8 @@ def test_works
   def test_early_stopping
     x_train, y_train, x_test, y_test = regression_data
 
-    model = XGBoost::Regressor.new
-    model.fit(x_train, y_train, early_stopping_rounds: 5, eval_set: [[x_test, y_test]], verbose: false)
+    model = XGBoost::Regressor.new(early_stopping_rounds: 5)
+    model.fit(x_train, y_train, eval_set: [[x_test, y_test]], verbose: false)
     assert_equal 9, model.booster.best_iteration
   end