数字识别

比赛说明

• MNIST（"修改后的国家标准与技术研究所"）是计算机视觉的事实上的 "hello world" 数据集。自1999年发布以来，手写图像的经典数据集已成为基准分类算法的基础。随着新机器学习技术的出现，MNIST 仍然是研究人员和学习者的可靠资源。
• 在本次比赛中，您的目标是正确 识别数以万计手写图像的数字 。我们策划了一套教程式的内核，涵盖从回归到神经网络的一切。我们鼓励您尝试使用不同的算法来学习第一手什么是有效的，以及技术如何比较。

注意：项目规范

成员角色

角色	用户	内容	代码
负责: knn	诺木人	knn项目文档	knn项目代码
负责: svm	马小穆	svm项目文档	svm项目代码
负责: 随机森林	平淡的天	随机森林项目文档	随机森林项目代码
负责: 神经网络	平淡的天	神经网络项目文档	神经网络项目代码
负责: cnn	==	cnn项目文档	cnn项目代码

数字识别 第一期(2018-04-18)

组长	组员	组员	组员	组员	组员	组员
限定心态	strengthen	VS53MV	不会修电脑	远心	小耀哥_0011	丨

数字识别 第二期(2018-04-21)

组长	组员	组员	组员	组员	组员
凌少skier	Blue	Max	考拉	Happyorg	过客

数字识别 第三期(2018-05-03)

负责人	组员	组员
技术负责人-诺木人 辅助负责人-平淡的心 辅助负责人-张凯 活动发起人-片刻	ifeng draw Faith ggggggggo 嘿！漆漆 kickfilper Lucien Chen L~Q~W 琴剑蓝天 時間ｄāń漠 歲寒✅已认证 給力小青年 星尘	瑛瑛wang 有一个人很酷 静水流深 ♡稳稳的幸福 Verestràsz vslyu :) 菠菜 QQ小冰 浅紫色 R ROOT

数字识别 第四期(2018-05-08)

负责人	组员	组员	组员
技术负责人-诺木人 辅助负责人-BrianCai 辅助负责人-嘿！漆漆	兰博归来 柳生 ZARD Forever 你别理我我没网 666 黄蛟 冬冬 荼蘼 烁今	简雨 B0lt1st nickine dying in the sun 王琪琪 常想一二 以朱代墨 Mang0 TonyZERO	冰花小子 阿铮 zh哲 小菜鸡 电酱prpr 琉璃火 张假飞 HAN Shuai 有人@我 天儿 Jaybo

开发流程

• 分类问题：0～9 数字
• 常用算法：knn、决策树、朴素贝叶斯、Logistic回归、SVM、集成方法（随机森林和 AdaBoost）

步骤:
一. 数据分析
1. 下载并加载数据
2. 总体预览数据:了解每列数据的含义,数据的格式等
3. 数据初步分析,使用统计学与绘图: 由于特征没有特殊的含义，不需要过多的细致分析

二. 特征工程
1.根据业务,常识,以及第二步的数据分析构造特征工程.
2.将特征转换为模型可以辨别的类型(如处理缺失值,处理文本进行等)

三. 模型选择
1.根据目标函数确定学习类型,是无监督学习还是监督学习,是分类问题还是回归问题等.
2.比较各个模型的分数,然后取效果较好的模型作为基础模型.

四. 模型融合
跳过，这个项目的重点是让大家都了解这个kaggle比赛怎么和算法更好的融合在一起。

五. 修改特征和模型参数
此处不做过多分析，主要是优化各个算法的参数。

* KNN => k值
* SVM => 惩罚系数，核函数
* RF => 树个数，树深度，叶子数
* PCA => 特征数 or 信息熵

一. 数据分析

数据下载和加载

数据集下载地址：https://www.kaggle.com/c/digit-recognizer/data

import os
import csv
import datetime
import numpy as np
import pandas as pd
from sklearn.decomposition import PCA
from sklearn.neighbors import KNeighborsClassifier


data_dir = '/opt/data/kaggle/getting-started/digit-recognizer/'


# 加载数据
def opencsv():
    # 使用 pandas 打开
    data = pd.read_csv(os.path.join(data_dir, 'input/train.csv'))
    data1 = pd.read_csv(os.path.join(data_dir, 'input/test.csv'))

    train_data = data.values[0:, 1:]  # 读入全部训练数据,  [行，列]
    train_label = data.values[0:, 0]  # 读取列表的第一列
    test_data = data1.values[0:, 0:]  # 测试全部测试个数据
    return train_data, train_label, test_data


# 加载数据
trainData, trainLabel, testData = opencsv()

总体预览数据(目标变量+数据特征)

• label: 目标变量（分类标签）
• pixel0~pixel783: 数据特征（分类属性），特征之间没有特别的业务联系（所以没必要进行统计分析了）

label,pixel0,pixel1,pixel2,pixel3,pixel4,pixel5,pixel6,pixel7,pixel8,pixel9,pixel10,pixel11,pixel12,pixel13,pixel14,pixel15,pixel16,pixel17,pixel18,pixel19,pixel20,pixel21,pixel22,pixel23,pixel24,pixel25,pixel26,pixel27,pixel28,pixel29,pixel30,pixel31,pixel32,pixel33,pixel34,pixel35,pixel36,pixel37,pixel38,pixel39,pixel40,pixel41,pixel42,pixel43,pixel44,pixel45,pixel46,pixel47,pixel48,pixel49,pixel50,pixel51,pixel52,pixel53,pixel54,pixel55,pixel56,pixel57,pixel58,pixel59,pixel60,pixel61,pixel62,pixel63,pixel64,pixel65,pixel66,pixel67,pixel68,pixel69,pixel70,pixel71,pixel72,pixel73,pixel74,pixel75,pixel76,pixel77,pixel78,pixel79,pixel80,pixel81,pixel82,pixel83,pixel84,pixel85,pixel86,pixel87,pixel88,pixel89,pixel90,pixel91,pixel92,pixel93,pixel94,pixel95,pixel96,pixel97,pixel98,pixel99,pixel100,pixel101,pixel102,pixel103,pixel104,pixel105,pixel106,pixel107,pixel108,pixel109,pixel110,pixel111,pixel112,pixel113,pixel114,pixel115,pixel116,pixel117,pixel118,pixel119,pixel120,pixel121,pixel122,pixel123,pixel124,pixel125,pixel126,pixel127,pixel128,pixel129,pixel130,pixel131,pixel132,pixel133,pixel134,pixel135,pixel136,pixel137,pixel138,pixel139,pixel140,pixel141,pixel142,pixel143,pixel144,pixel145,pixel146,pixel147,pixel148,pixel149,pixel150,pixel151,pixel152,pixel153,pixel154,pixel155,pixel156,pixel157,pixel158,pixel159,pixel160,pixel161,pixel162,pixel163,pixel164,pixel165,pixel166,pixel167,pixel168,pixel169,pixel170,pixel171,pixel172,pixel173,pixel174,pixel175,pixel176,pixel177,pixel178,pixel179,pixel180,pixel181,pixel182,pixel183,pixel184,pixel185,pixel186,pixel187,pixel188,pixel189,pixel190,pixel191,pixel192,pixel193,pixel194,pixel195,pixel196,pixel197,pixel198,pixel199,pixel200,pixel201,pixel202,pixel203,pixel204,pixel205,pixel206,pixel207,pixel208,pixel209,pixel210,pixel211,pixel212,pixel213,pixel214,pixel215,pixel216,pixel217,pixel218,pixel219,pixel220,pixel221,pixel222,pixel223,pixel224,pixel225,pixel226,pixel227,pixel228,pixel229,pixel230,pixel231,pixel232,pixel233,pixel234,pixel235,pixel236,pixel237,pixel238,pixel239,pixel240,pixel241,pixel242,pixel243,pixel244,pixel245,pixel246,pixel247,pixel248,pixel249,pixel250,pixel251,pixel252,pixel253,pixel254,pixel255,pixel256,pixel257,pixel258,pixel259,pixel260,pixel261,pixel262,pixel263,pixel264,pixel265,pixel266,pixel267,pixel268,pixel269,pixel270,pixel271,pixel272,pixel273,pixel274,pixel275,pixel276,pixel277,pixel278,pixel279,pixel280,pixel281,pixel282,pixel283,pixel284,pixel285,pixel286,pixel287,pixel288,pixel289,pixel290,pixel291,pixel292,pixel293,pixel294,pixel295,pixel296,pixel297,pixel298,pixel299,pixel300,pixel301,pixel302,pixel303,pixel304,pixel305,pixel306,pixel307,pixel308,pixel309,pixel310,pixel311,pixel312,pixel313,pixel314,pixel315,pixel316,pixel317,pixel318,pixel319,pixel320,pixel321,pixel322,pixel323,pixel324,pixel325,pixel326,pixel327,pixel328,pixel329,pixel330,pixel331,pixel332,pixel333,pixel334,pixel335,pixel336,pixel337,pixel338,pixel339,pixel340,pixel341,pixel342,pixel343,pixel344,pixel345,pixel346,pixel347,pixel348,pixel349,pixel350,pixel351,pixel352,pixel353,pixel354,pixel355,pixel356,pixel357,pixel358,pixel359,pixel360,pixel361,pixel362,pixel363,pixel364,pixel365,pixel366,pixel367,pixel368,pixel369,pixel370,pixel371,pixel372,pixel373,pixel374,pixel375,pixel376,pixel377,pixel378,pixel379,pixel380,pixel381,pixel382,pixel383,pixel384,pixel385,pixel386,pixel387,pixel388,pixel389,pixel390,pixel391,pixel392,pixel393,pixel394,pixel395,pixel396,pixel397,pixel398,pixel399,pixel400,pixel401,pixel402,pixel403,pixel404,pixel405,pixel406,pixel407,pixel408,pixel409,pixel410,pixel411,pixel412,pixel413,pixel414,pixel415,pixel416,pixel417,pixel418,pixel419,pixel420,pixel421,pixel422,pixel423,pixel424,pixel425,pixel426,pixel427,pixel428,pixel429,pixel430,pixel431,pixel432,pixel433,pixel434,pixel435,pixel436,pixel437,pixel438,pixel439,pixel440,pixel441,pixel442,pixel443,pixel444,pixel445,pixel446,pixel447,pixel448,pixel449,pixel450,pixel451,pixel452,pixel453,pixel454,pixel455,pixel456,pixel457,pixel458,pixel459,pixel460,pixel461,pixel462,pixel463,pixel464,pixel465,pixel466,pixel467,pixel468,pixel469,pixel470,pixel471,pixel472,pixel473,pixel474,pixel475,pixel476,pixel477,pixel478,pixel479,pixel480,pixel481,pixel482,pixel483,pixel484,pixel485,pixel486,pixel487,pixel488,pixel489,pixel490,pixel491,pixel492,pixel493,pixel494,pixel495,pixel496,pixel497,pixel498,pixel499,pixel500,pixel501,pixel502,pixel503,pixel504,pixel505,pixel506,pixel507,pixel508,pixel509,pixel510,pixel511,pixel512,pixel513,pixel514,pixel515,pixel516,pixel517,pixel518,pixel519,pixel520,pixel521,pixel522,pixel523,pixel524,pixel525,pixel526,pixel527,pixel528,pixel529,pixel530,pixel531,pixel532,pixel533,pixel534,pixel535,pixel536,pixel537,pixel538,pixel539,pixel540,pixel541,pixel542,pixel543,pixel544,pixel545,pixel546,pixel547,pixel548,pixel549,pixel550,pixel551,pixel552,pixel553,pixel554,pixel555,pixel556,pixel557,pixel558,pixel559,pixel560,pixel561,pixel562,pixel563,pixel564,pixel565,pixel566,pixel567,pixel568,pixel569,pixel570,pixel571,pixel572,pixel573,pixel574,pixel575,pixel576,pixel577,pixel578,pixel579,pixel580,pixel581,pixel582,pixel583,pixel584,pixel585,pixel586,pixel587,pixel588,pixel589,pixel590,pixel591,pixel592,pixel593,pixel594,pixel595,pixel596,pixel597,pixel598,pixel599,pixel600,pixel601,pixel602,pixel603,pixel604,pixel605,pixel606,pixel607,pixel608,pixel609,pixel610,pixel611,pixel612,pixel613,pixel614,pixel615,pixel616,pixel617,pixel618,pixel619,pixel620,pixel621,pixel622,pixel623,pixel624,pixel625,pixel626,pixel627,pixel628,pixel629,pixel630,pixel631,pixel632,pixel633,pixel634,pixel635,pixel636,pixel637,pixel638,pixel639,pixel640,pixel641,pixel642,pixel643,pixel644,pixel645,pixel646,pixel647,pixel648,pixel649,pixel650,pixel651,pixel652,pixel653,pixel654,pixel655,pixel656,pixel657,pixel658,pixel659,pixel660,pixel661,pixel662,pixel663,pixel664,pixel665,pixel666,pixel667,pixel668,pixel669,pixel670,pixel671,pixel672,pixel673,pixel674,pixel675,pixel676,pixel677,pixel678,pixel679,pixel680,pixel681,pixel682,pixel683,pixel684,pixel685,pixel686,pixel687,pixel688,pixel689,pixel690,pixel691,pixel692,pixel693,pixel694,pixel695,pixel696,pixel697,pixel698,pixel699,pixel700,pixel701,pixel702,pixel703,pixel704,pixel705,pixel706,pixel707,pixel708,pixel709,pixel710,pixel711,pixel712,pixel713,pixel714,pixel715,pixel716,pixel717,pixel718,pixel719,pixel720,pixel721,pixel722,pixel723,pixel724,pixel725,pixel726,pixel727,pixel728,pixel729,pixel730,pixel731,pixel732,pixel733,pixel734,pixel735,pixel736,pixel737,pixel738,pixel739,pixel740,pixel741,pixel742,pixel743,pixel744,pixel745,pixel746,pixel747,pixel748,pixel749,pixel750,pixel751,pixel752,pixel753,pixel754,pixel755,pixel756,pixel757,pixel758,pixel759,pixel760,pixel761,pixel762,pixel763,pixel764,pixel765,pixel766,pixel767,pixel768,pixel769,pixel770,pixel771,pixel772,pixel773,pixel774,pixel775,pixel776,pixel777,pixel778,pixel779,pixel780,pixel781,pixel782,pixel783
1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,3,141,139,3,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,9,254,254,8,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,9,254,254,8,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,9,254,254,106,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,9,254,254,184,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,9,254,254,184,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,9,254,254,184,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,6,185,254,184,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,89,254,184,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,4,146,254,184,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,9,254,254,184,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,9,254,254,184,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,9,254,254,184,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,9,254,254,184,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,9,254,254,184,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,156,254,254,184,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,185,255,255,184,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,185,254,254,184,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,185,254,254,184,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,63,254,254,62,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
4,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,220,179,6,0,0,0,0,0,0,0,0,9,77,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,28,247,17,0,0,0,0,0,0,0,0,27,202,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,242,155,0,0,0,0,0,0,0,0,27,254,63,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,160,207,6,0,0,0,0,0,0,0,27,254,65,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,127,254,21,0,0,0,0,0,0,0,20,239,65,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,77,254,21,0,0,0,0,0,0,0,0,195,65,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,70,254,21,0,0,0,0,0,0,0,0,195,142,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,56,251,21,0,0,0,0,0,0,0,0,195,227,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,222,153,5,0,0,0,0,0,0,0,120,240,13,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,67,251,40,0,0,0,0,0,0,0,94,255,69,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,234,184,0,0,0,0,0,0,0,19,245,69,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,234,169,0,0,0,0,0,0,0,3,199,182,10,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,154,205,4,0,0,26,72,128,203,208,254,254,131,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,61,254,129,113,186,245,251,189,75,56,136,254,73,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,15,216,233,233,159,104,52,0,0,0,38,254,73,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,18,254,73,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,18,254,73,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,5,206,106,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,186,159,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,6,209,101,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,25,130,155,254,254,254,157,30,2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,8,103,253,253,253,253,253,253,253,253,114,2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,11,208,253,253,253,253,253,253,253,253,253,253,107,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,31,253,253,253,253,253,253,253,253,253,253,253,215,101,3,0,0,0,0,0,0,0,0,0,0,0,0,23,210,253,253,253,248,161,222,222,246,253,253,253,253,253,39,0,0,0,0,0,0,0,0,0,0,0,0,136,253,253,253,229,77,0,0,0,70,218,253,253,253,253,215,91,0,0,0,0,0,0,0,0,0,0,5,214,253,253,253,195,0,0,0,0,0,104,224,253,253,253,253,215,29,0,0,0,0,0,0,0,0,0,116,253,253,253,247,75,0,0,0,0,0,0,26,200,253,253,253,253,216,4,0,0,0,0,0,0,0,0,254,253,253,253,195,0,0,0,0,0,0,0,0,26,200,253,253,253,253,5,0,0,0,0,0,0,0,0,254,253,253,253,99,0,0,0,0,0,0,0,0,0,25,231,253,253,253,36,0,0,0,0,0,0,0,0,254,253,253,253,99,0,0,0,0,0,0,0,0,0,0,223,253,253,253,129,0,0,0,0,0,0,0,0,254,253,253,253,99,0,0,0,0,0,0,0,0,0,0,127,253,253,253,129,0,0,0,0,0,0,0,0,254,253,253,253,99,0,0,0,0,0,0,0,0,0,0,139,253,253,253,90,0,0,0,0,0,0,0,0,254,253,253,253,99,0,0,0,0,0,0,0,0,0,78,248,253,253,253,5,0,0,0,0,0,0,0,0,254,253,253,253,216,34,0,0,0,0,0,0,0,33,152,253,253,253,107,1,0,0,0,0,0,0,0,0,206,253,253,253,253,140,0,0,0,0,0,30,139,234,253,253,253,154,2,0,0,0,0,0,0,0,0,0,16,205,253,253,253,250,208,106,106,106,200,237,253,253,253,253,209,22,0,0,0,0,0,0,0,0,0,0,0,82,253,253,253,253,253,253,253,253,253,253,253,253,253,209,22,0,0,0,0,0,0,0,0,0,0,0,0,1,91,253,253,253,253,253,253,253,253,253,253,213,90,7,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,18,129,208,253,253,253,253,159,129,90,4,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0

二. 特征工程（特征之间关系不大，此步骤跳过）

1. 当然可以做一些归一化的操作
2. 做降维：因为有些列一直都为了0，信息熵几乎为0，没有使用的必要

# 数据预处理-降维 PCA主成成分分析
def dRPCA(x_train, x_test, COMPONENT_NUM):
    print('dimensionality reduction...')
    trainData = np.array(x_train)
    testData = np.array(x_test)
    '''
    使用说明：https://www.cnblogs.com/pinard/p/6243025.html
    n_components>=1
      n_components=NUM   设置占特征数量比
    0 < n_components < 1
      n_components=0.99  设置阈值总方差占比
    '''
    pca = PCA(n_components=COMPONENT_NUM, whiten=False)
    pca.fit(trainData)  # Fit the model with X
    pcaTrainData = pca.transform(trainData)  # Fit the model with X and 在X上完成降维.
    pcaTestData = pca.transform(testData)  # Fit the model with X and 在X上完成降维.

    # pca 方差大小、方差占比、特征数量
    # print("方差大小:\n", pca.explained_variance_, "方差占比:\n", pca.explained_variance_ratio_)
    print("特征数量: %s" % pca.n_components_)
    print("总方差占比: %s" % sum(pca.explained_variance_ratio_))
    return pcaTrainData, pcaTestData


# 降维处理
trainDataPCA, testDataPCA = dRPCA(trainData, testData, 0.8)

三. 模型选择

1. 根据目标函数确定学习类型,是无监督学习还是监督学习,是分类问题还是回归问题等.

2. 比较各个模型的分数,然后取效果较好的模型作为基础模型.

3. 分类问题：0～9 数字

4. 常用算法：knn、决策树、朴素贝叶斯、Logistic回归、SVM、集成方法（随机森林和 AdaBoost）

knn

def trainModel(trainData, trainLabel):
    clf = KNeighborsClassifier()  # default:k = 5,defined by yourself:KNeighborsClassifier(n_neighbors=10)
    clf.fit(trainData, np.ravel(trainLabel))  # ravel Return a contiguous flattened array.
    return clf


# 模型训练
clf = trainModel(trainDataPCA, trainLabel)
# 结果预测
testLabel = clf.predict(testDataPCA)

svm

# 训练模型
def trainModel(trainData, trainLabel):
    print('Train SVM...')
    clf = SVC(C=4, kernel='rbf')
    clf.fit(trainData, trainLabel)  # 训练SVM
    return clf


# 模型训练
clf = trainModel(trainDataPCA, y_train)
# 结果预测
testLabel = clf.predict(pcaTestData)

RF - Random Forest

# 训练模型
def trainModel(X_train, y_train):
    print('Train RF...')
    clf = RandomForestClassifier(
        n_estimators=10,
        max_depth=10,
        min_samples_split=2,
        min_samples_leaf=1,
        random_state=34)
    clf.fit(X_train, y_train)  # 训练rf
    return clf


# 模型训练
clf = trainModel(trainDataPCA, y_train)
# 结果预测
testLabel = clf.predict(pcaTestData)

结果导出

def saveResult(result, csvName):
    with open(csvName, 'wb') as myFile:
        myWriter = csv.writer(myFile)
        myWriter.writerow(["ImageId", "Label"])
        index = 0
        for i in result:
            tmp = []
            index = index+1
            tmp.append(index)
            # tmp.append(i)
            tmp.append(int(i))
            myWriter.writerow(tmp)

# 结果的输出
saveResult(testLabel, '/opt/data/kaggle/getting-started/digit-recognizer/output/Result_xxx.csv')

四. 模型融合

跳过，这个项目的重点是让大家都了解这个kaggle比赛怎么和算法更好的融合在一起。

五. 修改特征和模型参数

此处不做过多分析，主要是优化各个算法的参数。

• KNN => k值
• SVM => 惩罚系数，核函数
• RF => 树个数，树深度，叶子数
• PCA => 特征数 or 信息熵

---

我们一直在努力

apachecn/AiLearning

---