bobo老师机器学习笔记-第四课：KNN算法

自己参考Bobo老师写得代码：

主要分为四个文件： knn.py中实现KNN算法、model_selection.py封装了样本数据的一些工具方法，比如切分为训练集和测试集；

metrics用来对模型进行评估、client用来调用算法进行运行

# -*- encoding: utf-8 -*-
"""
实现KNN的分类算法
"""
import numpy as np
from math import sqrt
from collections import Counter
from metrics import accuracy_score


class KnnClassifier(object):
    """
    K-近邻算法,（K Nearest Neighbour），简称KNN
    """

    def __init__(self, k):
        """
        K表示
        :param k: 表示参考的个数
        """
        self.k = k

    def fit(self, X_train, y_train):
        """
        利用输入的样本集进行训练KNN算法
        :param X_train: X 训练样本集
        :param y_train: y
        :return:
        """
        self.X_train = X_train
        self.y_train = y_train

        return self

    def predict(self, x_test):
        """
        对于输入的测试样本x进行预测
        :param x_test: 这个一个行向量
        :return:
        """

        assert x_test.shape[1] == self.X_train.shape[1], u'预测样本和训练样本的特征值不相等'
        # step1 用欧几里得算法计算x与周边的距离
        pridect_list = []
        for one_x in x_test:
            distances = [sqrt(np.sum((x - one_x) ** 2)) for x in self.X_train]
            sorted_index = np.argsort(distances)
            fit_y = self.y_train[sorted_index[:self.k]]
            target_label = Counter(fit_y).most_common()[0][0]
            pridect_list.append(target_label)
        return np.asarray(pridect_list, dtype='int32')

    def scores(self, y_pridect, y_test):

        return accuracy_score(y_pridect, y_test)

    def __repr__(self):
        return 'knn(k=%s)' % self.k

Metrics文件：

# -*- encoding: utf-8 -*-
"""
这个文件主要是计算一些指标比如准确度，用来评估算法的好还
"""

import numpy as np

def accuracy_score(y_test, y_pridect):
    """
    用来计算准确度
    :param y_test: 样本的标记测试集和
    :param y_pridect: 预测集
    :return:
    """

    assert y_pridect.shape[0] == y_test.shape[0], u'测试集和预测集的数据个数不相等'

    cnt = np.sum(y_test==y_pridect)

    return cnt / len(y_pridect)

model_selection.py文件：

# -*- encoding: utf-8 -*-
"""
这个文件主要是计算一些指标比如准确度，用来评估算法的好还
"""

import numpy as np

def accuracy_score(y_test, y_pridect):
    """
    用来计算准确度
    :param y_test: 样本的标记测试集和
    :param y_pridect: 预测集
    :return:
    """

    assert y_pridect.shape[0] == y_test.shape[0], u'测试集和预测集的数据个数不相等'

    cnt = np.sum(y_test==y_pridect)

    return cnt / len(y_pridect)

client文件进行测试：

from knn import KnnClassifier

from sklearn import datasets
from model_selection import train_test_split
from metrics import accuracy_score
import numpy as np

if __name__ == '__main__':
    knn = KnnClassifier(3)

    iris = datasets.load_iris()
    x_train, y_train, x_test, y_test = train_test_split(iris.data, iris.target, 0.7)
    classifier = knn.fit(x_train, y_train)
    y_pridect = classifier.predict(x_test)
    print(classifier.scores(y_pridect, y_test))