- 思想极度简单
- 应用数学知识少(近乎为零)
- 效果好
K近邻算法
import numpy as np
from math import sqrt
from collections import Counter
from ml_utils.metrics import accuray_score
class kNN_classify:
def __init__(self,k):
"""初始化kNN分类器"""
assert k>=1,"k must be valid"
self.k = k
self._X_train = None
self._y_train = None
def fit(self,X_train,y_train):
"""根据训练数据集 X_train 和 y_train 训练kNN 分类器"""
assert X_train.shape[0] == y_train.shape[0], \
"the size of X_train must equal to size of y_train"
assert self.k <= X_train.shape[0], \
"the size of X_train must at least k."
self._X_train = X_train
self._y_train =y_train
return self
def predict(self,X_predict,weight='distance'):
"""给定预测数据集X—-predict ,返回表示X_predict的结果向量"""
assert self._X_train is not None and self._y_train is not None,\
"must fit before predict!"
assert X_predict.shape[1] == self._X_train.shape[1],\
"the feature number of X——predict must to X_train "
y_predict =[self._predict(x,weight) for x in X_predict]
return np.array(y_predict)
def _predict(self,x,weight):
"""给定单个待预测数据x,返回X的预测结果值"""
assert x.shape[0] == self._X_train.shape[1], \
"the feature number of x must to be equal X_train "
# 计算欧式距离
distances = [sqrt(sum((x_train - x) ** 2)) for x_train in self._X_train]
# argsort 给出从大到小排序索引
nearest = np.argsort(distances)
# 按索引筛选出 最近距离 所对应的k个点的ytrain分类
topk_y = [self._y_train[m] for m in nearest[:self.k]]
if(weight=='uniform'):
#基于权重 样本点距离越近则权重越大
top_distinct = [ 1/distances[m] for m in nearest[:self.k]]
distinct_byweight={}
for m in range(len(topk_y)):
key = [topk_y[m]][0]
if(key not in distinct_byweight):
distinct_byweight[key] = top_distinct[m]
else:
distinct_byweight[key] += top_distinct[m]
maxClasser=None
maxclass = 0.0
#计算最大距离权重的点
for m in distinct_byweight:
if(distinct_byweight[m]>maxclass):
maxclass = distinct_byweight[m]
maxClasser = m
return maxClasser
# 统计分类中的的元素
votes = Counter(topk_y)
# 返回第一多的元素 的 值
return votes.most_common(1)[0][0]
def score(self,X_test,y_test,weight='distance'):
"根据测试数据及测试 X_test 和 y_test 确定模型的准确度"
y_predict = self.predict(X_test,weight)
return accuray_score(y_predict,y_test)
def __repr__(self):
return "KNN(k = % d)" % self.k
KNN 核心 是使用 欧拉距离公式 二维平面上两点a(x1,y1)与b(x2,y2)间的欧欧氏距离:
(
x
1
−
y
1
)
2
+
(
x
2
−
y
2
)
2
\sqrt{(x_1-y_1)^2+(x_2-y_2)^2}
(x1−y1)2+(x2−y2)2=
∑
i
=
1
n
(
x
i
a
−
x
i
b
)
2
\sqrt{\sum_{i=1}^{n}{(x_i^a-x_i^b)^2}}
∑i=1n(xia−xib)2
曼哈顿距离:
∑
i
=
1
n
∣
X
i
a
−
X
i
b
∣
\sum_{i=1}^{n}|X^{a}_i-X^{b}_i|
∑i=1n∣Xia−Xib∣
明可夫斯基距离:
(
∑
i
=
1
n
∣
X
i
a
−
X
i
b
∣
p
)
1
/
p
(\sum_{i=1}^{n}|X^{a}_i-X^{b}_i|^p)^{1/p}
(∑i=1n∣Xia−Xib∣p)1/p
1.计算 数据集里每个样本点与 预测样本点 的 距离
distances = [sqrt(sum((x_train - x) ** 2)) for x_train in self._X_train]
2.找到最近的k个样本点的距离所对应的类别
# argsort 给出从大到小排序索引
nearest = np.argsort(distances)
# 按索引筛选出 最近距离 所对应的k个点的ytrain分类
topk_y = [self._y_train[m] for m in nearest[:self.k]]
统计出和自己最相似类别返回这个预测的类别
# 统计分类中的的元素
votes = Counter(topk_y)
# 返回第一多的元素 的 值
return votes.most_common(1)[0][0]
由于给定的数据可能是 按类别顺序排列的 并且只有一份数据 所以就需要把数据集做打乱 按比例分割 成 训练;训练数据 训练数据标签 测试数据 测试数据标签
def train_test_split(X,y,test_ratio = 0.2,seed =None):
"""将数据X和y 按照 test_ratio分割成X_train,X_test,y_train,y_test"""
assert X.shape[0] ==y.shape[0],\
"the size of X must be equal to size of y"
assert 0.0 <= test_ratio <= 1.0,\
"test_ration must be valid"
if seed:
#随机打乱
np.random.seed(seed)
#相当于 随机生成 训练数据集的索引
shuffle_indexes = np.random.permutation(len(X))
#按比例分割大小
test_size = int(len(X) * test_ratio)
#按一定比例 划分数据集的索引
test_indexes = shuffle_indexes[:test_size]
train_indexes = shuffle_indexes[test_size:]
#根据打乱的 索引返回对应的数据
x_train = X[train_indexes]
y_train = y[train_indexes]
x_test = X[test_indexes]
y_test = y[test_indexes]
return x_train,y_train,x_test,y_test
def accuray_score(y_true,y_predict):
'''计算y_true 和 y_predict之间的准确率'''
assert y_true.shape[0] == y_predict.shape[0],\
"the size of y_true must be equal to the size of y_predict"
return sum(y_true == y_predict) / len(y_true)
测试
if __name__ == '__main__':
digits = datasets.load_digits()
x = digits.data
y = digits.target
some_fdigit = x[777]
some_digit_image = some_fdigit.reshape(8, 8)
plt.imshow(some_digit_image, cmap=matplotlib.cm.binary)
plt.show()
x_train, y_train, x_test, y_test = train_test_split(x, y);
knnclassify = kNN_classify(k=3);
knnclassify.fit(x_train, y_train)
print(knnclassify.score(x_test,y_test))
如果
距离的倒数 距离越远权重越低
if(uniform):
#基于权重 样本点距离越近则权重越大
top_distinct = [ 1/distances[m] for m in nearest[:self.k]]
distinct_byweight={}
for m in range(len(topk_y)):
key = [topk_y[m]][0]
if(key not in distinct_byweight):
distinct_byweight[key] = top_distinct[m]
else:
distinct_byweight[key] += top_distinct[m]
maxClasser=None
maxclass = 0.0
#计算最大距离权重的点
for m in distinct_byweight:
if(distinct_byweight[m]>maxclass):
maxclass = distinct_byweight[m]
maxClasser = m
return maxClasser
调参(超参数)
- 明可夫斯基距离
- 基于权重
KNN能干嘛
knn简单,效果强大
- 解决分类问题(多分类)
- 解决回归问题(KNeighborsRegressor)
缺点:
- 效率低下(如果训练集有m个样本,n个特征,则预测每一个新的数据,需要O(m*n))
- 高度数据相关
- 预测结果不具有可解释性
- 维数灾难(随着纬度的增加,“看似相近”的两个点之间距离越来越大)
1维 | 0到1的距离 | 1 |
---|---|---|
2维 | (0,0)到(1,1)的距离 | 1.414 |
3维 | (0,0,0) 到 (1,1,1)的距离 | 1.73 |
64维 | (0,0…,0,0)到 (1,1…,1,1) | 8 |
10000维 | (0,0,0…,0,0)到 (1,1,1…,1,1) | 100 |