kNNpython实现

kNN.py:

# coding:utf-8import numpy as np
from math import sqrt
from collections import Counter
from metrics import accuracy_scoreclass kNNClassifier:def __init__(self, k):&＃39;&＃39;&＃39;初始化KNN分类器&＃39;&＃39;&＃39;assert 1 <&＃61; k, &＃39;k must be valid&＃39;self.k &＃61; kself._X_train &＃61; Noneself._y_train &＃61; Nonedef fit(self, X_train, y_train):&＃39;&＃39;&＃39;根据训练数据集X-train和y_train训练kNN分类器&＃39;&＃39;&＃39;assert X_train.shape[0] &＃61;&＃61; y_train.shape[0], &＃39;the size of X_train must equal to the size of y_train&＃39;assert self.k <&＃61; X_train.shape[0], &＃39;the size of X_train must be at least k&＃39;self._X_train &＃61; X_trainself._y_train &＃61; y_trainreturn selfdef predict(self, X_predict):&＃39;&＃39;&＃39;给定带预测数据集X_predict,返回表示X_predict的结果向量&＃39;&＃39;&＃39;assert self._X_train is not None and self._y_train is not None, &＃39;must fit before predict&＃39;assert X_predict.shape[1] &＃61;&＃61; self._X_train.shape[1], &＃39;the feature number of X_predict equal to the feature number of self._X_train&＃39;y_predict &＃61; [self._predict(x) for x in X_predict]return np.array(y_predict)def _predict(self, x):&＃39;&＃39;&＃39;给定单个待预测数据x&＃xff0c;返回x的预测结果值&＃39;&＃39;&＃39;assert x.shape[0] &＃61;&＃61; self._X_train.shape[1], &＃39;the feature number of x must be equal to X_train&＃39;distances &＃61; [sqrt(np.sum((x_train - x)**2))for x_train in self._X_train]nearest &＃61; np.argsort(distances)topK_y &＃61; [self._y_train[i] for i in nearest[:self.k]]votes &＃61; Counter(topK_y)return votes.most_common(1)[0][0]def score(self, X_test, y_test):&＃39;&＃39;&＃39;计算对测试集的预测准确度&＃39;&＃39;&＃39;y_predict &＃61; self.predict(X_test)return accuracy_score(y_test, y_predict)def __repr__(self):return &＃39;kNN(k&＃61;%d)&＃39; % self.k# 使用:
# kNN_clf &＃61; kNNClassifier(3)
# kNN_clf.fit(X_train,y_train)
# kNN_clf.predict(x)
# kNN_clf.score(X_test,y_test)


metrics.py:

import numpy as npdef accuracy_score(y_test,y_predict):&＃39;&＃39;&＃39;计算y_predict相对于y_test的准确度&＃39;&＃39;&＃39;assert y_test.shape[0] &＃61;&＃61; y_predict.shape[0],&＃39;the size of y_test must equal to the size of y_predict&＃39;return sum(y_predict &＃61;&＃61; y_test) / len(y_test)


数据归一化处理&＃xff08;均值方差归一化&＃xff09;&＃xff1a;

import numpy as np class StandardScaler:def __init__(self):self.mean_ &＃61; Noneself.scale_ &＃61; Nonedef fit(self,X):&＃39;&＃39;&＃39;根据训练数据集X获得数据均值和方差&＃39;&＃39;&＃39;assert X.ndim &＃61;&＃61; 2,&＃39;the dimension of X must be 2&＃39;self.mean_ &＃61; np.array([np.mean(X[:,i]) for i in range(X.shape[1])])self.scale_ &＃61; np.array([np.std(X[:,i]) for i in range(X.shape[1])])return selfdef transform(self,X):&＃39;&＃39;&＃39;将X根据已有信息进行均值方差归一化处理&＃39;&＃39;&＃39;assert X.ndim &＃61;&＃61; 2,&＃39;the dimension of X must be 2&＃39;assert self.mean_ is not None and self.scale_ is not None , &＃39;must fit before transform!&＃39;assert X.shape[1] &＃61;&＃61; len(self.mean_),&＃39;the feature number of X must be equal to mean_ and std_&＃39;resX &＃61; np.empty(shape &＃61; X.shape,dtype &＃61; float)for col in range(0,X.shape[1]):resX[:,col] &＃61; (X[:,col] - self.mean_[col])/self.scale_[col]return resX