【统计学习笔记】KNN

模型

算法

线性扫描计算耗时，当训练数据集大时效率很低。

代码实现

KNN

class KNN:
    def __init__(self, X_train, y_train, n_neighbors=3, p=2):
        """
        parameter: n_neighbors 临近点个数
        parameter: p 距离度量
        """
        self.n = n_neighbors
        self.p = p
        self.X_train = X_train
        self.y_train = y_train
    
    def predict(self, X):
        # 取出n个点
        knn_list = []
        for i in range(self.n):
            dist = np.linalg.norm(X - self.X_train[i], ord=self.p)
            knn_list.append((dist, self.y_train[i]))
            
        for i in range(self.n, len(self.X_train)):
            max_index = knn_list.index(max(knn_list, key=lambda x: x[0]))
            dist = np.linalg.norm(X - self.X_train[i], ord=self.p)
            if knn_list[max_index][0] > dist:
                knn_list[max_index] = (dist, self.y_train[i])
                
        # 统计
        knn = [k[-1] for k in knn_list]
        count_pairs = Counter(knn)
        max_count = sorted(count_pairs, key=lambda x:x)[-1]
        return max_count
    
    def score(self, X_test, y_test):
        right_count = 0
        n = 10
        for X, y in zip(X_test, y_test):
            label = self.predict(X)
            if label == y:
                right_count += 1
        return right_count / len(X_test)

kd树

# kd-tree每个结点中主要包含的数据结构如下 
class KdNode(object):
    def __init__(self, dom_elt, split, left, right):
        self.dom_elt = dom_elt  # k维向量节点(k维空间中的一个样本点)
        self.split = split      # 整数（进行分割维度的序号）
        self.left = left        # 该结点分割超平面左子空间构成的kd-tree
        self.right = right      # 该结点分割超平面右子空间构成的kd-tree
class KdTree(object):
    def __init__(self, data):
        k = len(data[0])  # 数据维度
        
        def CreateNode(split, data_set): # 按第split维划分数据集exset创建KdNode
            if not data_set:    # 数据集为空
                return None
            # key参数的值为一个函数，此函数只有一个参数且返回一个值用来进行比较
            # operator模块提供的itemgetter函数用于获取对象的哪些维的数据，参数为需要获取的数据在对象中的序号
            #data_set.sort(key=itemgetter(split)) # 按要进行分割的那一维数据排序
            data_set.sort(key=lambda x: x[split])
            split_pos = len(data_set) // 2      # //为Python中的整数除法
            median = data_set[split_pos]        # 中位数分割点             
            split_next = (split + 1) % k        # cycle coordinates
            
            # 递归的创建kd树
            return KdNode(median, split, 
                          CreateNode(split_next, data_set[:split_pos]),     # 创建左子树
                          CreateNode(split_next, data_set[split_pos + 1:])) # 创建右子树
                                
        self.root = CreateNode(0, data)         # 从第0维分量开始构建kd树,返回根节点
# KDTree的前序遍历
def preorder(root):  
    print (root.dom_elt)  
    if root.left:      # 节点不为空
        preorder(root.left)  
    if root.right:  
        preorder(root.right)      

# 对构建好的kd树进行搜索，寻找与目标点最近的样本点：
from math import sqrt
from collections import namedtuple
# 定义一个namedtuple,分别存放最近坐标点、最近距离和访问过的节点数
result = namedtuple("Result_tuple", "nearest_point  nearest_dist  nodes_visited")
def find_nearest(tree, point):
    k = len(point) # 数据维度
    def travel(kd_node, target, max_dist):
        if kd_node is None:     
            return result([0] * k, float("inf"), 0) # python中用float("inf")和float("-inf")表示正负无穷
        nodes_visited = 1
        s = kd_node.split        # 进行分割的维度
        pivot = kd_node.dom_elt  # 进行分割的“轴”
        if target[s] <= pivot[s]:           # 如果目标点第s维小于分割轴的对应值(目标离左子树更近)
            nearer_node  = kd_node.left     # 下一个访问节点为左子树根节点
            further_node = kd_node.right    # 同时记录下右子树
        else:                               # 目标离右子树更近
            nearer_node  = kd_node.right    # 下一个访问节点为右子树根节点
            further_node = kd_node.left
        temp1 = travel(nearer_node, target, max_dist)  # 进行遍历找到包含目标点的区域
        nearest = temp1.nearest_point       # 以此叶结点作为“当前最近点”
        dist = temp1.nearest_dist           # 更新最近距离
        nodes_visited += temp1.nodes_visited  
        if dist < max_dist:     
            max_dist = dist    # 最近点将在以目标点为球心，max_dist为半径的超球体内
        temp_dist = abs(pivot[s] - target[s])    # 第s维上目标点与分割超平面的距离
        if  max_dist < temp_dist:                # 判断超球体是否与超平面相交
            return result(nearest, dist, nodes_visited) # 不相交则可以直接返回，不用继续判断
        #----------------------------------------------------------------------  
        # 计算目标点与分割点的欧氏距离  
        temp_dist = sqrt(sum((p1 - p2) ** 2 for p1, p2 in zip(pivot, target)))     
        if temp_dist < dist:         # 如果“更近”
            nearest = pivot          # 更新最近点
            dist = temp_dist         # 更新最近距离
            max_dist = dist          # 更新超球体半径
        # 检查另一个子结点对应的区域是否有更近的点
        temp2 = travel(further_node, target, max_dist) 
        nodes_visited += temp2.nodes_visited
        if temp2.nearest_dist < dist:        # 如果另一个子结点内存在更近距离
            nearest = temp2.nearest_point    # 更新最近点
            dist = temp2.nearest_dist        # 更新最近距离
        return result(nearest, dist, nodes_visited)
    return travel(tree.root, point, float("inf"))  # 从根节点开始递归