3.李航机器学习-KNN公式笔记及实现（未完成）

K近邻分类算法

项目链接：https://github.com/Wchenguang/gglearn/blob/master/KNNClassifier-todo/李航机器学习讲解/KNNClassifier.ipynb

公式笔记

• Lp距离公式
  $$L_p\left(x_i,x_j\right)={\left(\sum _{l=1}^n{\left|x_i^{(l)}-x_j^{(l)}\right|}^p\right)}^{\frac{1}{p}}$$

• $ p=2 $ 为欧氏距离
  $$L_2\left(x_i,x_j\right)={\left(\sum _{l=1}^n{\left|x_i^{(l)}-x_j^{(l)}\right|}^2\right)}^{\frac{1}{2}}$$

• $ p=1 $ 为曼哈顿距离
  $$L_1\left(x_i,x_j\right)=\sum _{l=1}^n\left|x_i^{(l)}-x_j^{(l)}\right|$$

'''
kd树构造
'''
class kdTreeNode:
    def __init__(self, x, index, target_dim, parent):
        '''
        所有节点共用x
        利用索引来表示该节点所覆盖的数据
        '''
        self.x = x
        self.index = index
        self.target_dim = target_dim
        self.parent = parent
        self.left = None
        self.right = None
        self.target_index = None
    def set_left(self, left):
        self.left = left
    def set_right(self, right):
        self.right = right
    def set_target_index(self, target_index):
        self.target_index = target_index
class kdTree:
    def __init__(self, x):
        self.dim = len(x[0])
        self.x = x.copy()
        self.tree_done = False
    def _fit(self, node):
        if(node == None):
            return
        left_index, mid_index, right_index = self._get_sorted_index(node.index, node.target_dim)
        node.set_target_index(mid_index)
        #print(node.x[mid_index])
        if(len(left_index) != 0):
            #print('left')
            node.set_left(kdTreeNode(self.x, left_index, (node.target_dim + 1)%self.dim, node))
            self._fit(node.left)
        if(len(right_index) != 0):
            #print('right')
            node.set_right(kdTreeNode(self.x, right_index, (node.target_dim + 1)%self.dim, node))
            self._fit(node.right)
    def fit(self):
        self.root = kdTreeNode(self.x, np.arange(len(self.x)), 0, None)
        self._fit(self.root)
        return self
    def _get_sorted_index(self, index_list, dim):
        temp_val = np.hstack((self.x[index_list], np.array([index_list]).T))
        mid = len(temp_val) // 2
        temp_val = temp_val[temp_val[:,dim].argsort()]
        return temp_val[:mid][:, -1], temp_val[mid][-1], temp_val[mid+1:][:, -1]