1. (149, 5) 
2. 5.1  3.5  1.4  0.2  Iris-setosa
3. 0  4.9  3.0  1.4  0.2  Iris-setosa
4. 1  4.7  3.2  1.3  0.2  Iris-setosa
5. 2  4.6  3.1  1.5  0.2  Iris-setosa
6. 3  5.0  3.6  1.4  0.2  Iris-setosa
7. 4  5.4  3.9  1.7  0.4  Iris-setosa
8. (149, 5) 
9.     Sepal_Length  Sepal_Width  Petal_Length  Petal_Width            type
10. 0           4.5          2.3           1.3          0.3     Iris-setosa
11. 1           6.3          2.5           5.0          1.9  Iris-virginica
12. 2           5.1          3.4           1.5          0.2     Iris-setosa
13. 3           6.3          3.3           6.0          2.5  Iris-virginica
14. 4           6.8          3.2           5.9          2.3  Iris-virginica
15. 切分点： 29
16. label_classes: ['Iris-setosa', 'Iris-versicolor', 'Iris-virginica']
17. kNNDIY模型预测，基于原数据： 0.95
18. kNN模型预测，基于原数据预测： [0.96666667 1.         0.93333333 1.         0.93103448]
19. kNN模型预测，原数据PCA处理后： [1.         0.96       0.95918367]

1. class KNeighborsClassifier Found at: sklearn.neighbors._classification
2. 
3. class KNeighborsClassifier(NeighborsBase, KNeighborsMixin, 
4.     SupervisedIntegerMixin, ClassifierMixin):
5. """Classifier implementing the k-nearest neighbors vote.
6.     
7.     Read more in the :ref:`User Guide <classification>`.
8.     
9.     Parameters
10.     ----------
11.     n_neighbors : int, default=5
12.     Number of neighbors to use by default for :meth:`kneighbors` queries.
13.     
14.     weights : {'uniform', 'distance'} or callable, default='uniform'
15.     weight function used in prediction.  Possible values:
16.     
17.     - 'uniform' : uniform weights.  All points in each neighborhood
18.     are weighted equally.
19.     - 'distance' : weight points by the inverse of their distance.
20.     in this case, closer neighbors of a query point will have a
21.     greater influence than neighbors which are further away.
22.     - [callable] : a user-defined function which accepts an
23.     array of distances, and returns an array of the same shape
24.     containing the weights.
25.     
26.     algorithm : {'auto', 'ball_tree', 'kd_tree', 'brute'}, default='auto'
27.     Algorithm used to compute the nearest neighbors:
28.     
29.     - 'ball_tree' will use :class:`BallTree`
30.     - 'kd_tree' will use :class:`KDTree`
31.     - 'brute' will use a brute-force search.
32.     - 'auto' will attempt to decide the most appropriate algorithm
33.     based on the values passed to :meth:`fit` method.
34.     
35.     Note: fitting on sparse input will override the setting of
36.     this parameter, using brute force.
37.     
38.     leaf_size : int, default=30
39.     Leaf size passed to BallTree or KDTree.  This can affect the
40.     speed of the construction and query, as well as the memory
41.     required to store the tree.  The optimal value depends on the
42.     nature of the problem.
43.     
44.     p : int, default=2
45.     Power parameter for the Minkowski metric. When p = 1, this is
46.     equivalent to using manhattan_distance (l1), and euclidean_distance
47.     (l2) for p = 2. For arbitrary p, minkowski_distance (l_p) is used.
48.     
49.     metric : str or callable, default='minkowski'
50.     the distance metric to use for the tree.  The default metric is
51.     minkowski, and with p=2 is equivalent to the standard Euclidean
52.     metric. See the documentation of :class:`DistanceMetric` for a
53.     list of available metrics.
54.     If metric is "precomputed", X is assumed to be a distance matrix and
55.     must be square during fit. X may be a :term:`sparse graph`,
56.     in which case only "nonzero" elements may be considered neighbors.
57.     
58.     metric_params : dict, default=None
59.     Additional keyword arguments for the metric function.
60.     
61.     n_jobs : int, default=None
62.     The number of parallel jobs to run for neighbors search.
63.     ``None`` means 1 unless in a :obj:`joblib.parallel_backend` context.
64.     ``-1`` means using all processors. See :term:`Glossary <n_jobs>`
65.     for more details.
66.     Doesn't affect :meth:`fit` method.
67.     
68.     Attributes
69.     ----------
70.     classes_ : array of shape (n_classes,)
71.     Class labels known to the classifier
72.     
73.     effective_metric_ : str or callble
74.     The distance metric used. It will be same as the `metric` parameter
75.     or a synonym of it, e.g. 'euclidean' if the `metric` parameter set to
76.     'minkowski' and `p` parameter set to 2.
77.     
78.     effective_metric_params_ : dict
79.     Additional keyword arguments for the metric function. For most 
80.      metrics
81.     will be same with `metric_params` parameter, but may also contain the
82.     `p` parameter value if the `effective_metric_` attribute is set to
83.     'minkowski'.
84.     
85.     outputs_2d_ : bool
86.     False when `y`'s shape is (n_samples, ) or (n_samples, 1) during fit
87.     otherwise True.
88.     
89.     Examples
90.     --------
91.     >>> X = [[0], [1], [2], [3]]
92.     >>> y = [0, 0, 1, 1]
93.     >>> from sklearn.neighbors import KNeighborsClassifier
94.     >>> neigh = KNeighborsClassifier(n_neighbors=3)
95.     >>> neigh.fit(X, y)
96.     KNeighborsClassifier(...)
97.     >>> print(neigh.predict([[1.1]]))
98.     [0]
99.     >>> print(neigh.predict_proba([[0.9]]))
100.     [[0.66666667 0.33333333]]
101.     
102.     See also
103.     --------
104.     RadiusNeighborsClassifier
105.     KNeighborsRegressor
106.     RadiusNeighborsRegressor
107.     NearestNeighbors
108.     
109.     Notes
110.     -----
111.     See :ref:`Nearest Neighbors <neighbors>` in the online 
112.      documentation
113.     for a discussion of the choice of ``algorithm`` and ``leaf_size``.
114.     
115.     .. warning::
116.     
117.     Regarding the Nearest Neighbors algorithms, if it is found that two
118.     neighbors, neighbor `k+1` and `k`, have identical distances
119.     but different labels, the results will depend on the ordering of the
120.     training data.
121.     
122.     https://en.wikipedia.org/wiki/K-nearest_neighbor_algorithm
123.     """
124.     @_deprecate_positional_args
125. def __init__(self, n_neighbors=5, 
126.         *, weights='uniform', algorithm='auto', leaf_size=30, 
127.         p=2, metric='minkowski', metric_params=None, n_jobs=None, **
128.         kwargs):
129. super().__init__(n_neighbors=n_neighbors, algorithm=algorithm, 
130.          leaf_size=leaf_size, metric=metric, p=p, metric_params=metric_params, 
131.          n_jobs=n_jobs, **kwargs)
132.         self.weights = _check_weights(weights)
133. 
134. def predict(self, X):
135. """Predict the class labels for the provided data.
136. 
137.         Parameters
138.         ----------
139.         X : array-like of shape (n_queries, n_features), \
140.                 or (n_queries, n_indexed) if metric == 'precomputed'
141.             Test samples.
142. 
143.         Returns
144.         -------
145.         y : ndarray of shape (n_queries,) or (n_queries, n_outputs)
146.             Class labels for each data sample.
147.         """
148.         X = check_array(X, accept_sparse='csr')
149.         neigh_dist, neigh_ind = self.kneighbors(X)
150.         classes_ = self.classes_
151.         _y = self._y
152. if not self.outputs_2d_:
153.             _y = self._y.reshape((-1, 1))
154.             classes_ = [self.classes_]
155.         n_outputs = len(classes_)
156.         n_queries = _num_samples(X)
157.         weights = _get_weights(neigh_dist, self.weights)
158.         y_pred = np.empty((n_queries, n_outputs), dtype=classes_[0].
159.          dtype)
160. for k, classes_k in enumerate(classes_):
161. if weights is None:
162.                 mode, _ = stats.mode(_y[neigh_indk], axis=1)
163. else:
164.                 mode, _ = weighted_mode(_y[neigh_indk], weights, axis=1)
165.             mode = np.asarray(mode.ravel(), dtype=np.intp)
166.             y_pred[:k] = classes_k.take(mode)
167. 
168. if not self.outputs_2d_:
169.             y_pred = y_pred.ravel()
170. return y_pred
171. 
172. def predict_proba(self, X):
173. """Return probability estimates for the test data X.
174. 
175.         Parameters
176.         ----------
177.         X : array-like of shape (n_queries, n_features), \
178.                 or (n_queries, n_indexed) if metric == 'precomputed'
179.             Test samples.
180. 
181.         Returns
182.         -------
183.         p : ndarray of shape (n_queries, n_classes), or a list of n_outputs
184.             of such arrays if n_outputs > 1.
185.             The class probabilities of the input samples. Classes are ordered
186.             by lexicographic order.
187.         """
188.         X = check_array(X, accept_sparse='csr')
189.         neigh_dist, neigh_ind = self.kneighbors(X)
190.         classes_ = self.classes_
191.         _y = self._y
192. if not self.outputs_2d_:
193.             _y = self._y.reshape((-1, 1))
194.             classes_ = [self.classes_]
195.         n_queries = _num_samples(X)
196.         weights = _get_weights(neigh_dist, self.weights)
197. if weights is None:
198.             weights = np.ones_like(neigh_ind)
199.         all_rows = np.arange(X.shape[0])
200.         probabilities = []
201. for k, classes_k in enumerate(classes_):
202.             pred_labels = _y[:k][neigh_ind]
203.             proba_k = np.zeros((n_queries, classes_k.size))
204. # a simple ':' index doesn't work right
205. for i, idx in enumerate(pred_labels.T): # loop is O(n_neighbors)
206.                 proba_k[all_rowsidx] += weights[:i]
207. 
208. # normalize 'votes' into real [0,1] probabilities
209.             normalizer = proba_k.sum(axis=1)[:np.newaxis]
210.             normalizer[normalizer == 0.0] = 1.0
211.             proba_k /= normalizer
212.             probabilities.append(proba_k)
213. 
214. if not self.outputs_2d_:
215.             probabilities = probabilities[0]
216. return probabilities