add _validate_input method to handle creation of classes_ attribute o…

…n fit call

mlrose_hiive/neural/_nn_core.py

@@ -6,6 +6,7 @@
 
 import numpy as np
 from abc import abstractmethod
+from sklearn.preprocessing import LabelBinarizer
 from mlrose_hiive.algorithms.decay import GeomDecay
 from mlrose_hiive.algorithms.rhc import random_hill_climb
 from mlrose_hiive.algorithms.sa import simulated_annealing
@@ -107,6 +108,20 @@ def _validate(self):
             raise Exception("""Algorithm must be one of: 'random_hill_climb',
                     'simulated_annealing', 'genetic_alg',
                     'gradient_descent'.""")
+
+    def _validate_input(self, X, y):
+        """
+        Add _classes attribute based on classes present in y.
+        """
+
+        # Required for sk-learn 1.3+. Doesn't cause issues for lower versions.
+        # Copied from https://github.com/scikit-learn/scikit-learn/blob/5c4aa5d0d90ba66247d675d4c3fc2fdfba3c39ff/sklearn/neural_network/_multilayer_perceptron.py
+        # Note: no workaround found for multi-class labels, still doesn't work with f1 score.
+
+        if (not hasattr(self, "classes_")):
+            self._label_binarizer = LabelBinarizer()
+            self._label_binarizer.fit(y)
+            self.classes_ = self._label_binarizer.classes_
 
     def fit(self, X, y=None, init_weights=None):
         """Fit neural network to data.
@@ -126,6 +141,7 @@ def fit(self, X, y=None, init_weights=None):
             If :code:`None`, then a random state is used.
         """
         self._validate()
+        self._validate_input(X, y)
 
         X, y = self._format_x_y_data(X, y)