Kaggle入门赛之Digit Recognizer

题目大意：手写数字的识别。每个数字由28*28的像素矩阵表示，也就是784个像素点。每个像素点的值between 0 and 255。

思路：knn在数字识别方面表现比较好，因为特征维数过多，kd_tree比较慢，所以我采用的是基于ball_tree的knn。每个像素的值都归一化，非0值都变成1。

工具：py2.7，sklearn，pycharm

# -*- coding: utf-8 -*-
import csv
import numpy as ny
from sklearn.neighbors import KNeighborsClassifier


def to_int(list):
	n=len(list)
	for i in range(n):
		list[i]=int(list[i])
	return list


# 归一化
def normalize(array):
	n,m=array.shape
	for i in range(n):
		for j in range(m):
			if array[i,j]!=0:
				array[i,j]=1
	return array


# 读取训练集
def load_train_data():
	train_data=[]
	train_label=[]
	with open('E:\\data\\kaggle\\digit recognizer\\train.csv','rb') as file:
		lines=csv.reader(file)
		header=True
		for line in lines:
			if header:
				header=False
				continue
			train_label.append(int(line[0]))
			train_data.append(to_int(line[1:]))
	return normalize(ny.array(train_data)),ny.array(train_label)


# 读取测试集
def load_test_data():
	test_data=[]
	with open('E:\\data\\kaggle\\digit recognizer\\test.csv','rb') as file:
		lines=csv.reader(file)
		header=True
		for line in lines:
			if header:
				header=False
				continue
			test_data.append(to_int(line))
	return normalize(ny.array(test_data))


def classify():
	train_data,train_label=load_train_data()
	test_data=load_test_data()
	neigh=KNeighborsClassifier(algorithm='ball_tree')
	neigh.fit(train_data,train_label)
	result=[]
	result.append(('ImageId','Label'))
	i=1
	for item in test_data:
		label=neigh.predict(ny.array(item).reshape((1,-1)))
		result.append((i,label[0]))
		i+=1
	with open('E:\\data\\kaggle\\digit recognizer\\result.csv','wb') as file:
		writer=csv.writer(file)
		writer.writerows(result)


classify()

大概跑了半小时就出来了，这是我的分数