fix the pickle error when dumping an estimator

.gitignore

@@ -129,4 +129,7 @@ dmypy.json
 .pyre/
 
 #Pycharm
-.idea
+.idea
+
+#VSCode
+.vscode

sklearn_genetic/genetic_search.py

@@ -257,7 +257,7 @@ def __init__(
         self.pre_dispatch = pre_dispatch
         self.error_score = error_score
         self.return_train_score = return_train_score
-        self.creator = creator
+        # self.creator = creator
         self.log_config = log_config
 
         # Check that the estimator is compatible with scikit-learn
@@ -299,8 +299,8 @@ def _register(self):
         """
         self.toolbox = base.Toolbox()
 
-        self.creator.create("FitnessMax", base.Fitness, weights=[self.criteria_sign])
-        self.creator.create("Individual", list, fitness=creator.FitnessMax)
+        creator.create("FitnessMax", base.Fitness, weights=[self.criteria_sign])
+        creator.create("Individual", list, fitness=creator.FitnessMax)
 
         attributes = []
         # Assign all the parameters defined in the param_grid
@@ -342,7 +342,7 @@ def _register(self):
         self._pop = self.toolbox.population(n=self.population_size)
         self._hof = tools.HallOfFame(self.keep_top_k)
 
-        self._stats = tools.Statistics(lambda ind: ind.fitness.values)
+        self._stats = tools.Statistics(getter_fittness_values)
         self._stats.register("fitness", np.mean)
         self._stats.register("fitness_std", np.std)
         self._stats.register("fitness_max", np.max)
@@ -540,8 +540,8 @@ def fit(self, X, y, callbacks=None):
             for k in range(len(self._hof))
         }
 
-        del self.creator.FitnessMax
-        del self.creator.Individual
+        del creator.FitnessMax
+        del creator.Individual
 
         return self
 
@@ -878,7 +878,7 @@ def __init__(
         self.pre_dispatch = pre_dispatch
         self.error_score = error_score
         self.return_train_score = return_train_score
-        self.creator = creator
+        # self.creator = creator
         self.log_config = log_config
 
         # Check that the estimator is compatible with scikit-learn
@@ -902,8 +902,8 @@ def _register(self):
 
         # Criteria sign to set max or min problem
         # And -1.0 as second weight to minimize number of features
-        self.creator.create("FitnessMax", base.Fitness, weights=[self.criteria_sign, -1.0])
-        self.creator.create("Individual", list, fitness=creator.FitnessMax)
+        creator.create("FitnessMax", base.Fitness, weights=[self.criteria_sign, -1.0])
+        creator.create("Individual", list, fitness=creator.FitnessMax)
 
         # Register the array to choose the features
         # Each binary value represents if the feature is selected or not
@@ -933,7 +933,7 @@ def _register(self):
 
         # Stats among axis 0 to get two values:
         # One based on the score and the other in the number of features
-        self._stats = tools.Statistics(lambda ind: ind.fitness.values)
+        self._stats = tools.Statistics(getter_fittness_values)
         self._stats.register("fitness", np.mean, axis=0)
         self._stats.register("fitness_std", np.std, axis=0)
         self._stats.register("fitness_max", np.max, axis=0)
@@ -1104,8 +1104,8 @@ def fit(self, X, y, callbacks=None):
 
         self.hof = self._hof
 
-        del self.creator.FitnessMax
-        del self.creator.Individual
+        del creator.FitnessMax
+        del creator.Individual
 
         return self
 
@@ -1365,3 +1365,10 @@ def score(self, X, y):
             ``best_estimator_.score`` method otherwise.
         """
         return self.estimator.score(self.transform(X), y)
+
+
+# helpers
+
+
+def getter_fittness_values(ind):
+    return ind.fitness.values

sklearn_genetic/tests/test_serialization.py

@@ -0,0 +1,105 @@
+import pytest
+from sklearn.datasets import load_iris, load_diabetes
+from sklearn.linear_model import SGDClassifier
+from sklearn.tree import DecisionTreeClassifier, DecisionTreeRegressor
+from sklearn.model_selection import train_test_split
+from sklearn.utils.validation import check_is_fitted
+from sklearn.cluster import KMeans
+from sklearn.metrics import accuracy_score, balanced_accuracy_score
+from sklearn.metrics import make_scorer
+import numpy as np
+
+from .. import GAFeatureSelectionCV
+from ..callbacks import (
+    ThresholdStopping,
+    DeltaThreshold,
+    ConsecutiveStopping,
+    TimerStopping,
+    ProgressBar,
+)
+from ..schedules import ExponentialAdapter, InverseAdapter
+from joblib import dump, load
+import os
+
+
+data = load_iris()
+label_names = data["target_names"]
+y = data["target"]
+X = data["data"]
+
+noise = np.random.uniform(1, 4, size=(X.shape[0], 10))
+
+X = np.hstack((X, noise))
+
+X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)
+
+
+def test_estimator_serialization():
+    clf = SGDClassifier(loss="modified_huber", fit_intercept=True)
+    generations = 6
+    evolved_estimator = GAFeatureSelectionCV(
+        clf,
+        cv=3,
+        scoring="accuracy",
+        population_size=6,
+        generations=generations,
+        tournament_size=3,
+        elitism=False,
+        keep_top_k=4,
+        verbose=False,
+        algorithm="eaSimple",
+        n_jobs=-1,
+        return_train_score=True,
+    )
+
+    evolved_estimator.fit(X_train, y_train)
+    dump_file = "evolved_estimator.pkl"
+
+    # test dump
+    assert dump(evolved_estimator, dump_file)[0] == dump_file
+
+    # load
+    dumped_estimator = load(dump_file)
+    features = dumped_estimator.support_
+
+    assert check_is_fitted(dumped_estimator) is None
+    assert features.shape[0] == X.shape[1]
+    assert len(dumped_estimator) == generations + 1  # +1 random initial population
+    assert len(dumped_estimator.predict(X_test)) == len(X_test)
+    assert dumped_estimator.score(X_train, y_train) >= 0
+    assert len(dumped_estimator.decision_function(X_test)) == len(X_test)
+    assert len(dumped_estimator.predict_proba(X_test)) == len(X_test)
+    assert len(dumped_estimator.predict_log_proba(X_test)) == len(X_test)
+    assert dumped_estimator.score(X_test, y_test) == accuracy_score(
+        y_test, dumped_estimator.predict(X_test)
+    )
+    assert bool(dumped_estimator.get_params())
+    assert len(dumped_estimator.hof) == dumped_estimator.keep_top_k
+    assert "gen" in dumped_estimator[0]
+    assert "fitness_max" in dumped_estimator[0]
+    assert "fitness" in dumped_estimator[0]
+    assert "fitness_std" in dumped_estimator[0]
+    assert "fitness_min" in dumped_estimator[0]
+
+    cv_results_ = dumped_estimator.cv_results_
+    cv_result_keys = set(cv_results_.keys())
+
+    assert "split0_test_score" in cv_result_keys
+    assert "split1_test_score" in cv_result_keys
+    assert "split2_test_score" in cv_result_keys
+    assert "split0_train_score" in cv_result_keys
+    assert "split1_train_score" in cv_result_keys
+    assert "split2_train_score" in cv_result_keys
+    assert "mean_test_score" in cv_result_keys
+    assert "std_test_score" in cv_result_keys
+    assert "rank_test_score" in cv_result_keys
+    assert "mean_train_score" in cv_result_keys
+    assert "std_train_score" in cv_result_keys
+    assert "rank_train_score" in cv_result_keys
+    assert "std_fit_time" in cv_result_keys
+    assert "mean_score_time" in cv_result_keys
+    assert "rank_n_features" in cv_result_keys
+    assert "features" in cv_result_keys
+
+    # delete dumped estimator
+    os.remove(dump_file)