kaggle预测房价
kaggle房价预测比赛官方地址:https://www.kaggle.com/c/house-prices-advanced-regression-techniques
kaggle数据集描述:https://www.kaggle.com/c/house-prices-advanced-regression-techniques/data
Step 1:引入相关的包
-
# coding:utf-8
-
# 注意读取文件时,Windows系统的\\和Linux系统的/的区别
-
import numpy as np
-
import pandas as pd
-
import matplotlib.pyplot as plt
-
from sklearn.linear_model import Ridge
-
from sklearn.model_selection import cross_val_score
-
from sklearn.ensemble import RandomForestRegressor
Step 2:读取数据
文件的组织形式是house price文件夹下面放house_price.py和input文件夹。input文件夹下面放的是从https://www.kaggle.com/c/house-prices-advanced-regression-techniques/data
下载的train.csv test.csv sample_submission.csv 和 data_description.txt 四个文件。
-
# 将csv数据转换为DataFrame数据,方便用pandas进行数据预处理
-
# 注意将print的注释打开,可以查看输出结果
-
#不要让pandas自己给数据加编号,这样ID就成为index了
-
train_df = pd.read_csv(".\\input\\train.csv",index_col = 0)
-
test_df = pd.read_csv('.\\input\\test.csv',index_col = 0)
-
# print train_df.shape
-
# print test_df.shape
-
# print train_df.head() # 默认展示前五行 这里是5行,80列
-
# print test_df.head() # 这里是5行,79列
Step 3:合并数据 :特征工程的工作!!!
这么做主要是为了用DF进行数据预处理的时候更加方便。等所有的需要的预处理进行完之后,我们再把他们分隔开。实际项目中,不会这样做。首先,SalePrice作为我们的训练目标,只会出现在训练集中,不会在测试集中。所以,我们先把SalePrice这一列给拿出来,不让它碍事儿。
# 看SalePrice的形状和用log1p处理后的形状
-
%matplotlib inline
-
prices = pd.DataFrame({'price':train_df['SalePrice'],'log(price+1)':np.log1p(train_df['SalePrice'])})
-
ps = prices.hist()
-
# plt.plot()
-
# plt.show()
-
# log1p即log(1+x),可以让label平滑化,将数据变为正态分布,目的在于使数据的呈现方式接近我们所希望的前提假设,从而进行更好的统计推断
-
y_train = np.log1p(train_df.pop('SalePrice')) #提出和test数据不一致的price,马上进行train和test的合并
-
all_df = pd.concat((train_df,test_df),axis = 0) #合并
-
# print all_df.shape #查看all_df (2919,79)
-
# print y_train.head() #查看处理后的标记预测值
Step 4:变量转化:特征工程和数据清洗的工作!!!
正确化变量属性:MSSubClass 的值其实应该是一个category(等级的划分),虽然是数字,但是代表多类别,Pandas是不会懂这些。使用DF的时候,这类数字符号会被默认记成数字。这种东西就很有误导性,我们需要把它变回成string
-
print all_df['MSSubClass'].dtypes #dtype('int64')
-
all_df['MSSubClass'] = all_df['MSSubClass'].astype(str) #转为string,便于查看他的分布情况
-
print all_df['MSSubClass'].dtypes
-
print all_df['MSSubClass'].value_counts()
把category的变量转变成numerical表达形式:当我们用numerical来表达categorical的时候,要注意,数字本身有大小的含义,所以乱用数字会给之后的模型学习带来麻烦。于是我们可以用One-Hot的方法来表达category。pandas自带的get_dummies方法,可以帮你一键做到One-Hot。
#不能被计算机理解的变量(字符串,离散型变量等)
-
print pd.get_dummies(all_df['MSSubClass'],prefix = 'MSSubClass'#处理离散型变量的方法get_dummies,即就是one-hot).head()
-
all_dummy_df = pd.get_dummies(all_df) #pandas自动选择那些事离散型变量,省去了我们做选择
-
print all_dummy_df.head()
清洗第二步:处理numerical变量:
比如,有一些数据是缺失的
-
print all_dummy_df.isnull().sum().sort_values(ascending = False).head(11) #查看缺失情况,按照缺失情况排序
-
# 注意:处理缺失情况时要看数据描述,确实值得处理方式工具意义和缺失情况有很大不同,有时确实本身就有意义,我们要把他当
#做一个类型,其他时候要将其补上或者删除这个特征
-
#我们这里用mean填充
-
mean_cols = all_dummy_df.mean()
-
print mean_cols.head(10)
-
all_dummy_df = all_dummy_df.fillna(mean_cols) #fillna填充
-
print all_dummy_df.isnull().sum().sum() #输出0
标准化numerical数据:
这一步并不是必要,但是得看你想要用的分类器是什么。一般来说,regression的分类器都需要这一步,最好是把源数据给放在一个标准分布内,不要让数据间的差距太大。我们不需要把One-Hot的那些0/1数据给标准化,因为只有0和1,我们的目标应该是那些本来就是numerical的数据型的特征。
-
numeric_cols = all_df.columns[all_df.dtypes != 'object'] #查看那些是numerical数据,本来就是数字化的数据
-
print numeric_cols
-
#标准化numerical数据,让数据更加平滑,更加便于计算:如z-score标准化:(x-x’)/s 【x:原数据;x':平均数;s:标准差】
-
numeric_col_means = all_dummy_df.loc[:,numeric_cols].mean() #均值
-
numeric_col_std = all_dummy_df.loc[:,numeric_cols].std() #标准差
-
all_dummy_df.loc[:,numeric_cols] = (all_dummy_df.loc[:,numeric_cols] - numeric_col_means) / numeric_col_std
Step 5-1: 建立模型【房价预测/Ridge/RandomForest/cross_validation】
-
# 把数据处理之后,分回训练集和测试集(起初在数据处理时将train和test数据结合了)
-
dummy_train_df = all_dummy_df.loc[train_df.index]
-
dummy_test_df = all_dummy_df.loc[test_df.index]
-
print dummy_train_df.shape,dummy_test_df.shape #输出((1460,303),(1459,303))
-
# 将DF数据转换成Numpy Array的形式,更好地配合sklearn
-
X_train = dummy_train_df.values
-
X_test = dummy_test_df.values
Ridge Regression(回归模型的一种:对于多因子的数据集,可以直接把所有的特征都放进去,不用考虑特征提取)
from sklearn.linear_model import Ridge
from sklearn.model_selection import cross_val_score #交叉验证来测试模型
#不是很必要,知识吧DataFrame转换成Numpy Array格式数据
X_train = dummy_train_df.values
X_test = dummy_test_df.values
#用Sklearn自带的cross_calidation来测试模型
-
alphas = np.logspace(-3,2,50) #创建等比梳理与,如:10^-3至10^2其中的50个数
-
test_scores = [] #交叉验证的得分,最后找到最好的参数
-
for alpha in alphas:
-
clf = Ridge(alpha)
-
test_score = np.sqrt(-cross_val_score(clf,X_train,y_train,cv = 10,scoring = 'neg_mean_squared_error'))
-
test_scores.append(np.mean(test_score))
plt.plot(alphas,test_scores) #可视化参数与分数
plt.title('Alpha vs CV Error')
-
plt.show()
- 存下所有的cv值,看看那个alpha值更好【调参数】
大概alpha=10~20的时候,可以把score达到0.135左右。
Random Forest
from sklearn.ensemble import RandomForestRegressorRF
-
max_features = [.1,.3,.5,.7,.9,.99]
-
test_scores = []
-
for max_feat in max_features:
-
clf = RandomForestRegressor(n_estimators = 200,max_features = max_feat)
-
test_score = np.sqrt(-cross_val_score(clf,X_train,y_train,cv = 5,scoring = 'neg_mean_squared_error'))
-
test_scores.append(np.mean(test_score))
-
plt.plot(max_features,test_scores)
-
plt.title('Max Features vs CV Error')
-
plt.show()
max_features=0.3时,RF达到了最优0.137
Step 5-2: 建立模型 【进阶版/bagging/boosting/AdaBoost/XGBoost】
从模型的角度考虑,用了bagging、boosting(AdaBoost)、XGBoost三个模型(模型框架)。
把数据集分回 训练/测试集
-
dummy_train_df = all_dummy_df.loc[train_df.index]
-
dummy_test_df = all_dummy_df.loc[test_df.index]
-
print dummy_train_df.shape,dummy_test_df.shape
-
# 将DF数据转换成Numpy Array的形式,更好地配合sklearn
-
X_train = dummy_train_df.values
-
X_test = dummy_test_df.values
1、bagging:
单个分类器的效果真的是很有限。我们会倾向于把N多的分类器合在一起,做一个“综合分类器”以达到最好的效果。我们从刚刚的试验中得知,Ridge(alpha=15)给了我们最好的结果
-
ridge = Ridge(alpha=15)
-
# bagging 把很多小的分类器放在一起,每个train随机的一部分数据,然后把它们的最终结果综合起来(多数投票)
-
# bagging 算是一种算法框架
-
params = [1, 10, 15, 20, 25, 30, 40] # 多少个弱分类器
-
test_scores = []
-
for param in params:
-
clf = BaggingRegressor(n_estimators=param,base_estimator = ridge) # #base_estimator = ridge是弱分类器0.132(params=25时)
-
#clf = BaggingRegressor(n_estimators = param)#用Bagging自带的DecisionTree,最好0.140
-
test_score = np.sqrt(-cross_val_score(clf, X_train, y_train, cv=10, scoring='neg_mean_squared_error'))
-
test_scores.append(np.mean(test_score))
-
plt.plot(params, test_scores)
-
plt.title('n_estimators vs CV Error')
-
plt.show()
-
br = BaggingRegressor(base_estimator=ridge, n_estimators=25)
-
br.fit(X_train, y_train)
-
y_final = np.expm1(br.predict(X_test))
2、boosting
Boosting比Bagging理论上更高级点,它也是揽来一把的分类器。但是把他们线性排列。下一个分类器把上一个分类器分类得不好的地方加上更高的权重,这样下一个分类器就能在这个部分学得更加“深刻”。
-
from sklearn.ensemble import AdaBoostRegressor
-
ms = [10,15,20,25,30,35,40,45,50]
-
test_scores = []
-
for param in params:
-
clf = AdaBoostRegressor(base_estimator = ridge,n_estimators = param) #ms=25时,0.132,但是不稳定,需要更多的参数或者更多小分类器
-
test_score = np.sqrt(-cross_val_score(clf,X_train,y_train,cv = 10,scoring = 'neg_mean_squared_error'))
-
test_scores.append(np.mean(test_score))
-
plt.plot(params,test_scores)
-
plt.title('n_estimators vs CV Error')
-
plt.show()
3、XGBoost (kaggle神器)
这依旧是一款Boosting框架的模型,但是却做了很多的改进。
-
from xgboost import XGBRegressor
-
params = [1,2,3,4,5,6]
-
test_scores = []
-
for param in params:
-
clf = XGBRegressor(max_depth = param) #深度params=5时,错误率达到0.127
-
test_score = np.sqrt(-cross_val_score(clf,X_train,y_train,cv = 10,scoring = 'neg_mean_squared_error'))
-
test_scores.append(np.mean(test_score))
-
plt.plot(params,test_scores)
-
plt.title('max_depth vs CV Error')
-
plt.show()
-
xgb = XGBRegressor(max_depth = 5)
-
xgb.fit(X_train, y_train)
-
y_final = np.expm1(xgb.predict(X_test))
Step 6: Ensemble
这里我们用一个Stacking的思维来汲取两种或者多种模型的优点 ;
首先,我们把最好的parameter拿出来,做成我们最终的model;
-
ridge = Ridge(alpha = 15)
-
rf = RandomForestRegressor(n_estimators = 500,max_features = .3)
-
ridge.fit(X_train,y_train)
-
rf.fit(X_train,y_train)
#最前面个label做了一个log(1+x),这里需要把predit的值给exp回去,并且戒掉那个‘1’
-
y_ridge = np.expm1(ridge.predict(X_test))
-
y_rf = np.expm1(rf.predict(X_test))
-
#把所有的model的预测结果作为新的输入,最简单的就是不下直接【平均化】
-
y_final = (y_ridge + y_rf) / 2
Step 7: 提交结果
注意提交的格式!包括大小写、索引、列头等小细节。
-
submission_df = pd.DataFrame(data = {'Id':test_df.index,'SalePrice':y_final})
-
print submission_df.head(10)
-
submission_df.to_csv('.\\input\\submission.csv',columns = ['Id','SalePrice'],index = False)
Step5-1版完整练习代码:
-
# coding:utf-8
-
# 注意Windows系统的\\和Linux系统的/的区别
-
import numpy as np
-
import pandas as pd
-
import matplotlib.pyplot as plt
-
from sklearn.linear_model import Ridge
-
from sklearn.model_selection import cross_val_score
-
from sklearn.ensemble import RandomForestRegressor
-
# 文件的组织形式是house price文件夹下面放house_price.py和input文件夹
-
# input文件夹下面放的是从https://www.kaggle.com/c/house-prices-advanced-regression-techniques/data下载的train.csv test.csv sample_submission.csv 和 data_description.txt 四个文件
-
# step1 检查源数据集,读入数据,将csv数据转换为DataFrame数据
-
train_df = pd.read_csv(".\\input\\train.csv",index_col = 0)
-
test_df = pd.read_csv('.\\input\\test.csv',index_col = 0)
-
# print train_df.shape
-
# print test_df.shape
-
# print train_df.head() # 默认展示前五行 这里是5行,80列
-
# print test_df.head() # 这里是5行,79列
-
# step2 合并数据,进行数据预处理
-
prices = pd.DataFrame({'price':train_df['SalePrice'],'log(price+1)':np.log1p(train_df['SalePrice'])})
-
# ps = prices.hist()
-
# plt.plot()
-
# plt.show()
-
y_train = np.log1p(train_df.pop('SalePrice'))
-
all_df = pd.concat((train_df,test_df),axis = 0)
-
# print all_df.shape
-
# print y_train.head()
-
# step3 变量转化
-
print all_df['MSSubClass'].dtypes
-
all_df['MSSubClass'] = all_df['MSSubClass'].astype(str)
-
print all_df['MSSubClass'].dtypes
-
print all_df['MSSubClass'].value_counts()
-
# 把category的变量转变成numerical表达形式
-
# get_dummies方法可以帮你一键one-hot
-
print pd.get_dummies(all_df['MSSubClass'],prefix = 'MSSubClass').head()
-
all_dummy_df = pd.get_dummies(all_df)
-
print all_dummy_df.head()
-
# 处理好numerical变量
-
print all_dummy_df.isnull().sum().sort_values(ascending = False).head(11)
-
# 我们这里用mean填充
-
mean_cols = all_dummy_df.mean()
-
print mean_cols.head(10)
-
all_dummy_df = all_dummy_df.fillna(mean_cols)
-
print all_dummy_df.isnull().sum().sum()
-
# 标准化numerical数据
-
numeric_cols = all_df.columns[all_df.dtypes != 'object']
-
print numeric_cols
-
numeric_col_means = all_dummy_df.loc[:,numeric_cols].mean()
-
numeric_col_std = all_dummy_df.loc[:,numeric_cols].std()
-
all_dummy_df.loc[:,numeric_cols] = (all_dummy_df.loc[:,numeric_cols] - numeric_col_means) / numeric_col_std
-
# step4 建立模型
-
# 把数据处理之后,送回训练集和测试集
-
dummy_train_df = all_dummy_df.loc[train_df.index]
-
dummy_test_df = all_dummy_df.loc[test_df.index]
-
print dummy_train_df.shape,dummy_test_df.shape
-
# 将DF数据转换成Numpy Array的形式,更好地配合sklearn
-
X_train = dummy_train_df.values
-
X_test = dummy_test_df.values
-
# Ridge Regression
-
# alphas = np.logspace(-3,2,50)
-
# test_scores = []
-
# for alpha in alphas:
-
# clf = Ridge(alpha)
-
# test_score = np.sqrt(-cross_val_score(clf,X_train,y_train,cv = 10,scoring = 'neg_mean_squared_error'))
-
# test_scores.append(np.mean(test_score))
-
# plt.plot(alphas,test_scores)
-
# plt.title('Alpha vs CV Error')
-
# plt.show()
-
# random forest
-
# max_features = [.1,.3,.5,.7,.9,.99]
-
# test_scores = []
-
# for max_feat in max_features:
-
# clf = RandomForestRegressor(n_estimators = 200,max_features = max_feat)
-
# test_score = np.sqrt(-cross_val_score(clf,X_train,y_train,cv = 5,scoring = 'neg_mean_squared_error'))
-
# test_scores.append(np.mean(test_score))
-
# plt.plot(max_features,test_scores)
-
# plt.title('Max Features vs CV Error')
-
# plt.show()
-
# Step 5: ensemble
-
# 用stacking的思维来汲取两种或者多种模型的优点
-
ridge = Ridge(alpha = 15)
-
rf = RandomForestRegressor(n_estimators = 500,max_features = .3)
-
ridge.fit(X_train,y_train)
-
rf.fit(X_train,y_train)
-
y_ridge = np.expm1(ridge.predict(X_test))
-
y_rf = np.expm1(rf.predict(X_test))
-
y_final = (y_ridge + y_rf) / 2
-
# Step 6: 提交结果
-
submission_df = pd.DataFrame(data = {'Id':test_df.index,'SalePrice':y_final})
-
print submission_df.head(10)
-
submission_df.to_csv('.\\input\\submission.csv',columns = ['Id','SalePrice'],index =
Step5-2版完整练习代码:
-
# coding:utf-8
-
# 注意Windows系统的\\和Linux系统的/的区别
-
import numpy as np
-
import pandas as pd
-
import matplotlib.pyplot as plt
-
from sklearn.linear_model import Ridge
-
from sklearn.model_selection import cross_val_score
-
from sklearn.ensemble import RandomForestRegressor
-
from sklearn.ensemble import BaggingRegressor
-
from sklearn.ensemble import AdaBoostRegressor
-
from xgboost import XGBRegressor
-
# 文件的组织形式是house price文件夹下面放house_price.py和input文件夹
-
# input文件夹下面放的是从https://www.kaggle.com/c/house-prices-advanced-regression-techniques/data下载的train.csv test.csv sample_submission.csv 和 data_description.txt 四个文件
-
# step1 检查源数据集,读入数据,将csv数据转换为DataFrame数据
-
train_df = pd.read_csv("./input/train.csv",index_col = 0)
-
test_df = pd.read_csv('./input/test.csv',index_col = 0)
-
# print train_df.shape
-
# print test_df.shape
-
# print train_df.head() # 默认展示前五行 这里是5行,80列
-
# print test_df.head() # 这里是5行,79列
-
# step2 合并数据,进行数据预处理
-
prices = pd.DataFrame({'price':train_df['SalePrice'],'log(price+1)':np.log1p(train_df['SalePrice'])})
-
# ps = prices.hist()
-
# plt.plot()
-
# plt.show()
-
y_train = np.log1p(train_df.pop('SalePrice'))
-
all_df = pd.concat((train_df,test_df),axis = 0)
-
# print all_df.shape
-
# print y_train.head()
-
# step3 变量转化
-
print all_df['MSSubClass'].dtypes
-
all_df['MSSubClass'] = all_df['MSSubClass'].astype(str)
-
print all_df['MSSubClass'].dtypes
-
print all_df['MSSubClass'].value_counts()
-
# 把category的变量转变成numerical表达形式
-
# get_dummies方法可以帮你一键one-hot
-
print pd.get_dummies(all_df['MSSubClass'],prefix = 'MSSubClass').head()
-
all_dummy_df = pd.get_dummies(all_df)
-
print all_dummy_df.head()
-
# 处理好numerical变量
-
print all_dummy_df.isnull().sum().sort_values(ascending = False).head(11)
-
# 我们这里用mean填充
-
mean_cols = all_dummy_df.mean()
-
print mean_cols.head(10)
-
all_dummy_df = all_dummy_df.fillna(mean_cols)
-
print all_dummy_df.isnull().sum().sum()
-
# 标准化numerical数据
-
numeric_cols = all_df.columns[all_df.dtypes != 'object']
-
print numeric_cols
-
numeric_col_means = all_dummy_df.loc[:,numeric_cols].mean()
-
numeric_col_std = all_dummy_df.loc[:,numeric_cols].std()
-
all_dummy_df.loc[:,numeric_cols] = (all_dummy_df.loc[:,numeric_cols] - numeric_col_means) / numeric_col_std
-
# step4 建立模型
-
# 把数据处理之后,送回训练集和测试集
-
dummy_train_df = all_dummy_df.loc[train_df.index]
-
dummy_test_df = all_dummy_df.loc[test_df.index]
-
print dummy_train_df.shape,dummy_test_df.shape
-
# 将DF数据转换成Numpy Array的形式,更好地配合sklearn
-
X_train = dummy_train_df.values
-
X_test = dummy_test_df.values
-
# Ridge Regression
-
# alphas = np.logspace(-3,2,50)
-
# test_scores = []
-
# for alpha in alphas:
-
# clf = Ridge(alpha)
-
# test_score = np.sqrt(-cross_val_score(clf,X_train,y_train,cv = 10,scoring = 'neg_mean_squared_error'))
-
# test_scores.append(np.mean(test_score))
-
# plt.plot(alphas,test_scores)
-
# plt.title('Alpha vs CV Error')
-
# plt.show()
-
# random forest
-
# max_features = [.1,.3,.5,.7,.9,.99]
-
# test_scores = []
-
# for max_feat in max_features:
-
# clf = RandomForestRegressor(n_estimators = 200,max_features = max_feat)
-
# test_score = np.sqrt(-cross_val_score(clf,X_train,y_train,cv = 5,scoring = 'neg_mean_squared_error'))
-
# test_scores.append(np.mean(test_score))
-
# plt.plot(max_features,test_scores)
-
# plt.title('Max Features vs CV Error')
-
# plt.show()
-
# ensemble
-
# 用stacking的思维来汲取两种或者多种模型的优点
-
# ridge = Ridge(alpha = 15)
-
# rf = RandomForestRegressor(n_estimators = 500,max_features = .3)
-
# ridge.fit(X_train,y_train)
-
# rf.fit(X_train,y_train)
-
# y_ridge = np.expm1(ridge.predict(X_test))
-
# y_rf = np.expm1(rf.predict(X_test))
-
# y_final = (y_ridge + y_rf) / 2
-
# 做一点高级的ensemble
-
ridge = Ridge(alpha = 15)
-
# bagging 把很多小的分类器放在一起,每个train随机的一部分数据,然后把它们的最终结果综合起来(多数投票)
-
# bagging 算是一种算法框架
-
# params = [1,10,15,20,25,30,40]
-
# test_scores = []
-
# for param in params:
-
# clf = BaggingRegressor(base_estimator = ridge,n_estimators = param)
-
# test_score = np.sqrt(-cross_val_score(clf,X_train,y_train,cv = 10,scoring = 'neg_mean_squared_error'))
-
# test_scores.append(np.mean(test_score))
-
# plt.plot(params,test_scores)
-
# plt.title('n_estimators vs CV Error')
-
# plt.show()
-
# br = BaggingRegressor(base_estimator = ridge,n_estimators = 25)
-
# br.fit(X_train,y_train)
-
# y_final = np.expm1(br.predict(X_test))
-
# boosting 比bagging更高级,它是弄来一把分类器,把它们线性排列,下一个分类器把上一个分类器分类不好的地方加上更高的权重,这样,下一个分类器在这部分就能学习得更深刻
-
# params = [10,15,20,25,30,35,40,45,50]
-
# test_scores = []
-
# for param in params:
-
# clf = AdaBoostRegressor(base_estimator = ridge,n_estimators = param)
-
# test_score = np.sqrt(-cross_val_score(clf,X_train,y_train,cv = 10,scoring = 'neg_mean_squared_error'))
-
# test_scores.append(np.mean(test_score))
-
# plt.plot(params,test_scores)
-
# plt.title('n_estimators vs CV Error')
-
# plt.show()
-
# xgboost
-
params = [1,2,3,4,5,6]
-
test_scores = []
-
for param in params:
-
clf = XGBRegressor(max_depth = param)
-
test_score = np.sqrt(-cross_val_score(clf,X_train,y_train,cv = 10,scoring = 'neg_mean_squared_error'))
-
test_scores.append(np.mean(test_score))
-
plt.plot(params,test_scores)
-
plt.title('max_depth vs CV Error')
-
plt.show()
-
xgb = XGBRegressor(max_depth = 5)
-
xgb.fit(X_train, y_train)
-
y_final = np.expm1(xgb.predict(X_test))
-
# 提交结果
-
submission_df = pd.DataFrame(data = {'Id':test_df.index,'SalePrice':y_final})
-
print submission_df.head(10)
-
submission_df.to_csv('./input/submission_xgboosting.csv',columns = ['Id','SalePrice'
总结:
并不是所有的数据源都是整齐划一的x=[var1,var2,var3...]
方法:
【非标准-现实生活数据】-->降维、取特征、数字化表达(特征工程)-->【高维数据】
【文本数据】-->单词出现次数、单词出现频率、语义网络等(特征工程)-->【数据】
【图片数据】-->RGB点阵-->【数组】
【视频数据】-->分为【音轨】and【视频轨】-->【音轨:声波/语音识别】【视频轨:一维图片/图片识别】
可参考:
https://www.cnblogs.com/irenelin/p/7400388.html
http://blog.csdn.net/qilixuening/article/details/75153131
http://blog.csdn.net/qilixuening/article/details/75151026
http://blog.csdn.net/chris_lee_hehe/article/details/78700140