ML之kNNC:基于iris莺尾花数据集(PCA处理+三维散点图可视化)利用kNN算法实现分类预测

设计思路

ML之kNNC:基于iris莺尾花数据集(PCA处理+三维散点图可视化)利用kNN算法实现分类预测


输出结果

ML之kNNC:基于iris莺尾花数据集(PCA处理+三维散点图可视化)利用kNN算法实现分类预测

ML之kNNC:基于iris莺尾花数据集(PCA处理+三维散点图可视化)利用kNN算法实现分类预测

ML之kNNC:基于iris莺尾花数据集(PCA处理+三维散点图可视化)利用kNN算法实现分类预测

(149, 5)

   5.1  3.5  1.4  0.2  Iris-setosa

0  4.9  3.0  1.4  0.2  Iris-setosa

1  4.7  3.2  1.3  0.2  Iris-setosa

2  4.6  3.1  1.5  0.2  Iris-setosa

3  5.0  3.6  1.4  0.2  Iris-setosa

4  5.4  3.9  1.7  0.4  Iris-setosa

(149, 5)

   Sepal_Length  Sepal_Width  Petal_Length  Petal_Width            type

0           4.5          2.3           1.3          0.3     Iris-setosa

1           6.3          2.5           5.0          1.9  Iris-virginica

2           5.1          3.4           1.5          0.2     Iris-setosa

3           6.3          3.3           6.0          2.5  Iris-virginica

4           6.8          3.2           5.9          2.3  Iris-virginica

切分点: 29

label_classes: ['Iris-setosa', 'Iris-versicolor', 'Iris-virginica']

kNNDIY模型预测,基于原数据: 0.95

kNN模型预测,基于原数据预测: [0.96666667 1.         0.93333333 1.         0.93103448]

kNN模型预测,原数据PCA处理后: [1.         0.96       0.95918367]



核心代码

class KNeighborsClassifier Found at: sklearn.neighbors._classification

class KNeighborsClassifier(NeighborsBase, KNeighborsMixin,

   SupervisedIntegerMixin, ClassifierMixin):

   """Classifier implementing the k-nearest neighbors vote.

   

   Read more in the :ref:`User Guide <classification>`.

   

   Parameters

   ----------

   n_neighbors : int, default=5

   Number of neighbors to use by default for :meth:`kneighbors` queries.

   

   weights : {'uniform', 'distance'} or callable, default='uniform'

   weight function used in prediction.  Possible values:

   

   - 'uniform' : uniform weights.  All points in each neighborhood

   are weighted equally.

   - 'distance' : weight points by the inverse of their distance.

   in this case, closer neighbors of a query point will have a

   greater influence than neighbors which are further away.

   - [callable] : a user-defined function which accepts an

   array of distances, and returns an array of the same shape

   containing the weights.

   

   algorithm : {'auto', 'ball_tree', 'kd_tree', 'brute'}, default='auto'

   Algorithm used to compute the nearest neighbors:

   

   - 'ball_tree' will use :class:`BallTree`

   - 'kd_tree' will use :class:`KDTree`

   - 'brute' will use a brute-force search.

   - 'auto' will attempt to decide the most appropriate algorithm

   based on the values passed to :meth:`fit` method.

   

   Note: fitting on sparse input will override the setting of

   this parameter, using brute force.

   

   leaf_size : int, default=30

   Leaf size passed to BallTree or KDTree.  This can affect the

   speed of the construction and query, as well as the memory

   required to store the tree.  The optimal value depends on the

   nature of the problem.

   

   p : int, default=2

   Power parameter for the Minkowski metric. When p = 1, this is

   equivalent to using manhattan_distance (l1), and euclidean_distance

   (l2) for p = 2. For arbitrary p, minkowski_distance (l_p) is used.

   

   metric : str or callable, default='minkowski'

   the distance metric to use for the tree.  The default metric is

   minkowski, and with p=2 is equivalent to the standard Euclidean

   metric. See the documentation of :class:`DistanceMetric` for a

   list of available metrics.

   If metric is "precomputed", X is assumed to be a distance matrix and

   must be square during fit. X may be a :term:`sparse graph`,

   in which case only "nonzero" elements may be considered neighbors.

   

   metric_params : dict, default=None

   Additional keyword arguments for the metric function.

   

   n_jobs : int, default=None

   The number of parallel jobs to run for neighbors search.

   ``None`` means 1 unless in a :obj:`joblib.parallel_backend` context.

   ``-1`` means using all processors. See :term:`Glossary <n_jobs>`

   for more details.

   Doesn't affect :meth:`fit` method.

   

   Attributes

   ----------

   classes_ : array of shape (n_classes,)

   Class labels known to the classifier

   

   effective_metric_ : str or callble

   The distance metric used. It will be same as the `metric` parameter

   or a synonym of it, e.g. 'euclidean' if the `metric` parameter set to

   'minkowski' and `p` parameter set to 2.

   

   effective_metric_params_ : dict

   Additional keyword arguments for the metric function. For most

    metrics

   will be same with `metric_params` parameter, but may also contain the

   `p` parameter value if the `effective_metric_` attribute is set to

   'minkowski'.

   

   outputs_2d_ : bool

   False when `y`'s shape is (n_samples, ) or (n_samples, 1) during fit

   otherwise True.

   

   Examples

   --------

   >>> X = [[0], [1], [2], [3]]

   >>> y = [0, 0, 1, 1]

   >>> from sklearn.neighbors import KNeighborsClassifier

   >>> neigh = KNeighborsClassifier(n_neighbors=3)

   >>> neigh.fit(X, y)

   KNeighborsClassifier(...)

   >>> print(neigh.predict([[1.1]]))

   [0]

   >>> print(neigh.predict_proba([[0.9]]))

   [[0.66666667 0.33333333]]

   

   See also

   --------

   RadiusNeighborsClassifier

   KNeighborsRegressor

   RadiusNeighborsRegressor

   NearestNeighbors

   

   Notes

   -----

   See :ref:`Nearest Neighbors <neighbors>` in the online

    documentation

   for a discussion of the choice of ``algorithm`` and ``leaf_size``.

   

   .. warning::

   

   Regarding the Nearest Neighbors algorithms, if it is found that two

   neighbors, neighbor `k+1` and `k`, have identical distances

   but different labels, the results will depend on the ordering of the

   training data.

   

   https://en.wikipedia.org/wiki/K-nearest_neighbor_algorithm

   """

   @_deprecate_positional_args

   def __init__(self, n_neighbors=5,

       *, weights='uniform', algorithm='auto', leaf_size=30,

       p=2, metric='minkowski', metric_params=None, n_jobs=None, **

       kwargs):

       super().__init__(n_neighbors=n_neighbors, algorithm=algorithm,

        leaf_size=leaf_size, metric=metric, p=p, metric_params=metric_params,

        n_jobs=n_jobs, **kwargs)

       self.weights = _check_weights(weights)

   

   def predict(self, X):

       """Predict the class labels for the provided data.

       Parameters

       ----------

       X : array-like of shape (n_queries, n_features), \

               or (n_queries, n_indexed) if metric == 'precomputed'

           Test samples.

       Returns

       -------

       y : ndarray of shape (n_queries,) or (n_queries, n_outputs)

           Class labels for each data sample.

       """

       X = check_array(X, accept_sparse='csr')

       neigh_dist, neigh_ind = self.kneighbors(X)

       classes_ = self.classes_

       _y = self._y

       if not self.outputs_2d_:

           _y = self._y.reshape((-1, 1))

           classes_ = [self.classes_]

       n_outputs = len(classes_)

       n_queries = _num_samples(X)

       weights = _get_weights(neigh_dist, self.weights)

       y_pred = np.empty((n_queries, n_outputs), dtype=classes_[0].

        dtype)

       for k, classes_k in enumerate(classes_):

           if weights is None:

               mode, _ = stats.mode(_y[neigh_indk], axis=1)

           else:

               mode, _ = weighted_mode(_y[neigh_indk], weights, axis=1)

           mode = np.asarray(mode.ravel(), dtype=np.intp)

           y_pred[:k] = classes_k.take(mode)

       

       if not self.outputs_2d_:

           y_pred = y_pred.ravel()

       return y_pred

   

   def predict_proba(self, X):

       """Return probability estimates for the test data X.

       Parameters

       ----------

       X : array-like of shape (n_queries, n_features), \

               or (n_queries, n_indexed) if metric == 'precomputed'

           Test samples.

       Returns

       -------

       p : ndarray of shape (n_queries, n_classes), or a list of n_outputs

           of such arrays if n_outputs > 1.

           The class probabilities of the input samples. Classes are ordered

           by lexicographic order.

       """

       X = check_array(X, accept_sparse='csr')

       neigh_dist, neigh_ind = self.kneighbors(X)

       classes_ = self.classes_

       _y = self._y

       if not self.outputs_2d_:

           _y = self._y.reshape((-1, 1))

           classes_ = [self.classes_]

       n_queries = _num_samples(X)

       weights = _get_weights(neigh_dist, self.weights)

       if weights is None:

           weights = np.ones_like(neigh_ind)

       all_rows = np.arange(X.shape[0])

       probabilities = []

       for k, classes_k in enumerate(classes_):

           pred_labels = _y[:k][neigh_ind]

           proba_k = np.zeros((n_queries, classes_k.size))

           # a simple ':' index doesn't work right

           for i, idx in enumerate(pred_labels.T): # loop is O(n_neighbors)

               proba_k[all_rowsidx] += weights[:i]

           

           # normalize 'votes' into real [0,1] probabilities

           normalizer = proba_k.sum(axis=1)[:np.newaxis]

           normalizer[normalizer == 0.0] = 1.0

           proba_k /= normalizer

           probabilities.append(proba_k)

       

       if not self.outputs_2d_:

           probabilities = probabilities[0]

       return probabilities


上一篇:AI大事件 | 2020年AI界爆款论文汇总,GPT-3 get新技能


下一篇:自娱小程序--九宫格