Merge pull request #92 from rodrigo-arenas/0.8.1

Version 0.81 Release

docs/release_notes.rst

@@ -3,6 +3,17 @@ Release Notes
 
 Some notes on new features in various releases
 
+What's new in 0.8.1
+-------------------
+
+^^^^^^^^^
+Features:
+^^^^^^^^^
+
+* If the `max_features` parameter from :class:`~sklearn_genetic.GAFeatureSelectionCV` is set,
+  the initial population is now sampled giving more probability to solutions with less than `max_features` features.
+
+
 What's new in 0.8.0
 -------------------
 

sklearn_genetic/_version.py

@@ -1 +1 @@
-__version__ = "0.8.0"
+__version__ = "0.8.1"

sklearn_genetic/genetic_search.py

@@ -24,6 +24,7 @@
     create_gasearch_cv_results_,
     create_feature_selection_cv_results_,
 )
+from .utils.random import weighted_choice
 
 
 class GASearchCV(BaseSearchCV):
@@ -940,6 +941,7 @@ def __init__(
         self.n_features = None
         self.X_ = None
         self.y_ = None
+        self.features_proportion = None
         self.callbacks = None
         self.best_features_ = None
         self.best_estimator_ = None
@@ -999,7 +1001,11 @@ def _register(self):
 
         # Register the array to choose the features
         # Each binary value represents if the feature is selected or not
-        self.toolbox.register("features", random.randint, 0, 1)
+
+        if self.features_proportion:
+            self.toolbox.register("features", weighted_choice, self.features_proportion)
+        else:
+            self.toolbox.register("features", random.randint, 0, 1)
 
         self.toolbox.register(
             "individual",
@@ -1107,10 +1113,11 @@ def evaluate(self, individual):
         self.logbook.record(parameters=current_generation_features)
 
         # Penalize individuals with more features than the max_features parameter
+
         if self.max_features and (
             n_selected_features > self.max_features or n_selected_features == 0
         ):
-            score = -self.criteria_sign * 10000
+            score = -self.criteria_sign * 100000
 
         return [score, n_selected_features]
 
@@ -1136,6 +1143,9 @@ def fit(self, X, y, callbacks=None):
         self.X_, self.y_ = check_X_y(X, y)
         self.n_features = X.shape[1]
 
+        if self.max_features:
+            self.features_proportion = self.max_features/self.n_features
+
         # Make sure the callbacks are valid
         self.callbacks = check_callback(callbacks)
 

sklearn_genetic/utils/random.py

@@ -0,0 +1,20 @@
+import random
+
+
+def weighted_choice(weight):
+    """
+    Parameters
+    ----------
+    weight: float
+        Weight of choosing a chromosome
+
+    Returns
+    -------
+        Bool random (not uniform) choice
+    """
+
+    # This help to don't generate individuals of the same size on average
+    p = random.uniform(0, weight)
+    choice = random.choices([0, 1], [1-p, p])[0]
+
+    return choice