Up my solution

sklearn_questions.py

@@ -55,13 +55,12 @@
 
 from sklearn.model_selection import BaseCrossValidator
 
-from sklearn.utils.validation import check_X_y, check_is_fitted
-from sklearn.utils.validation import check_array
+from sklearn.utils.validation import check_is_fitted
 from sklearn.utils.multiclass import check_classification_targets
 from sklearn.metrics.pairwise import pairwise_distances
 
 
-class KNearestNeighbors(BaseEstimator, ClassifierMixin):
+class KNearestNeighbors(ClassifierMixin, BaseEstimator):
     """KNearestNeighbors classifier."""
 
     def __init__(self, n_neighbors=1):  # noqa: D107
@@ -82,6 +81,16 @@ def fit(self, X, y):
         self : instance of KNearestNeighbors
             The current instance of the classifier
         """
+        X, y = self._validate_data(X, y, accept_sparse=True,
+                                   multi_output=False)
+        check_classification_targets(y)
+        self.classes_ = np.unique(y)
+        self.n_features_in_ = X.shape[1]
+        if len(self.classes_) < 2:
+            raise ValueError("Only 1 class is present.")
+        self.X_train_ = X
+        self.y_train_ = y
+
         return self
 
     def predict(self, X):
@@ -97,7 +106,16 @@ def predict(self, X):
         y : ndarray, shape (n_test_samples,)
             Predicted class labels for each test data sample.
         """
-        y_pred = np.zeros(X.shape[0])
+        check_is_fitted(self, ['X_train_', 'y_train_'])
+        X = self._validate_data(X, accept_sparse=True, reset=False)
+        y_pred = np.zeros(X.shape[0], dtype=self.y_train_.dtype)
+        distances = pairwise_distances(X, self.X_train_, metric='euclidean')
+        nearest_i = np.argsort(distances, axis=1)[:, :self.n_neighbors]
+        nearest_l = self.y_train_[nearest_i]
+        for i, labels in enumerate(nearest_l):
+            unique_labels, counts = np.unique(labels, return_counts=True)
+            y_pred[i] = unique_labels[np.argmax(counts)]
+
         return y_pred
 
     def score(self, X, y):
@@ -115,7 +133,12 @@ def score(self, X, y):
         score : float
             Accuracy of the model computed for the (X, y) pairs.
         """
-        return 0.
+        check_is_fitted(self)
+        X = self._validate_data(X, accept_sparse=True, reset=False)
+        y = self._validate_data(y, ensure_2d=False, reset=False)
+        y_predict = self.predict(X)
+        acc = np.mean(y_predict == y)
+        return acc
 
 
 class MonthlySplit(BaseCrossValidator):
@@ -155,7 +178,19 @@ def get_n_splits(self, X, y=None, groups=None):
         n_splits : int
             The number of splits.
         """
-        return 0
+        X_cc = X.copy()
+        if self.time_col == 'index':
+            X_cc = X_cc.reset_index()
+
+        if not pd.api.types.is_datetime64_any_dtype(X_cc[self.time_col]):
+            raise ValueError(
+                f"The column '{self.time_col}' is not a datetime."
+            )
+
+        X_cc = X_cc.sort_values(by=self.time_col)
+        unique_months = X_cc[self.time_col].dt.to_period('M').unique()
+
+        return len(unique_months) - 1
 
     def split(self, X, y, groups=None):
         """Generate indices to split data into training and test set.
@@ -177,12 +212,15 @@ def split(self, X, y, groups=None):
         idx_test : ndarray
             The testing set indices for that split.
         """
+        X_cc = X.reset_index()
+        n_splits = self.get_n_splits(X_cc, y, groups)
+        X_grp = X_cc.sort_values(by=self.time_col).groupby(
+            pd.Grouper(key=self.time_col, freq="M"))
+        idxs = [group.index for _, group in X_grp]
 
-        n_samples = X.shape[0]
-        n_splits = self.get_n_splits(X, y, groups)
         for i in range(n_splits):
-            idx_train = range(n_samples)
-            idx_test = range(n_samples)
+            idx_train = list(idxs[i])
+            idx_test = list(idxs[i+1])
             yield (
                 idx_train, idx_test
             )