# -*- coding:utf-8 -*-
__author__ = 'yangxin_ryan'
from numpy import *
from os import listdir
from collections import Counter
import operator
"""
图片的输入为 32 * 32的转换为 1 * 1024的向量
"""
class DigitalRecognition(object):
def __init__(self):
print("Welcome, 手写数字识别算法!")
"""
1.距离计算
tile生成和训练样本对应的矩阵,并与训练样本求差
取平方
将矩阵的每一行相加
开方
根据距离从小到大的排序,并返回对应的索引位置
2.选择距离最小的k个值
3.排序并返回出现最多的那个类型
"""
def classify_1(self, in_x, data_set, labels, k):
data_set_size = data_set.shape[0]
diff_mat = tile(in_x, (data_set_size, 1)) - data_set
sq_diff_mat = diff_mat ** 2
sq_distances = sq_diff_mat.sum(axis=1)
distances = sq_distances ** 0.5
sorted_dist_indicies = distances.argsort()
#
class_count = {}
for i in range(k):
vote_i_label = labels[sorted_dist_indicies[i]]
class_count[vote_i_label] = class_count.get(vote_i_label, 0) + 1
sorted_class_count = sorted(class_count.items(), key=operator.itemgetter(1), reverse=True)
return sorted_class_count
"""
1.计算距离
2.k个最近的标签
3.出现次数最多的标签即为最终类别
"""
def classify_2(self, in_x, data_set, labels, k):
dist = np.sum((in_x - data_set) ** 2, axis=1) ** 0.5
k_labels = [labels[index] for index in dist.argsort()[0:k]]
label = Counter(k_labels).most_common(1)[0][0]
return label
def file_to_matrix(self, file_name):
fr = open(file_name)
number_of_lines = len(fr.readlines())
return_mat = zeros((number_of_lines, 3))
class_label_vector = []
fr = open(file_name)
index = 0
for line in fr.readlines():
line = line.strip()
list_from_line = line.split("\t")
return_mat[index, :] = list_from_line[0:3]
class_label_vector.append(int(list_from_line[-1]))
index += 1
return return_mat, class_label_vector
"""
将图片转换为向量
图片的输入为 32 * 32的,将图像转换为向量,该函数创建 1 * 1024 的Numpy数组
"""
def img_to_vector(self, file_name):
return_vector = zeros((1, 1024))
fr = open(file_name, 'r')
for i in range(32):
line_str = fr.readline()
for j in range(32):
return_vector[0, 32 * i + j] = int(line_str[j])
return return_vector
def run(self, train_file_path, test_file_path, k):
labels = []
training_file_list = listdir(train_file_path)
train_len = len(training_file_list)
training_mat = zeros((train_len, 1024))
for i in range(train_len):
file_name_str = training_file_list[i]
file_str = file_name_str.split(".")[0]
class_num_str = int(file_str.split("_")[0])
labels.append(class_num_str)
img_file = train_file_path + file_name_str
print(img_file)
training_mat[i] = self.img_to_vector(img_file)
test_file_list = listdir(test_file_path)
error_count = 0.0
test_len = len(test_file_list)
for i in range(test_len):
file_name_str = test_file_list[i]
file_str = file_name_str.split(".")[0]
class_num_str = int(file_str.split("_")[0])
test_file_img = test_file_path + file_name_str
vector_under_test = self.img_to_vector(test_file_img)
classifier_result = self.classify_1(vector_under_test, training_mat, labels, k)
if classifier_result != class_num_str:
error_count += 1.0
print("\nthe total number of errors is: %d" % error_count)
print("\nthe total error rate is: %f" % (error_count / float(test_len)))
if __name__ == '__main__':
digital_recognition = DigitalRecognition()
digital_recognition.run("/Users/yangxin_ryan/PycharmProjects/MachineLearning/data/knn/trainingDigits/",
"/Users/yangxin_ryan/PycharmProjects/MachineLearning/data/knn/testDigits/",
3)