Python运用决策树方法和随机预测一个人的收入水平

数据集来源：

http://archive.ics.uci.edu/ml/datasets/Adult
该数据集是美国1994年人口普查数据库，预测任务是确定一个人的年收入是否超过5万美元。该数据集有以下指标，并且有缺失项。

步骤：

要对数据进行处理，删除缺失数据，然后对特征进行赋值，变换成哑变量。
举一个例子：

Edu	值
Preschool	1.0
1st-4th	2.0
5th-6th	3.0
7th-8th	4.0
9th	5.0
10th	6.0
11th	7.0
12th	8.0
HS-grad	9.0
Some-college	10.0
Assoc-acdm	11.0
Assoc-voc	12.0
Bachelors	13.0
Masters	14.0
Doctorate	15.0

方法：

决策树方法：这里用的是CART算法，max_dept最大深度取得过大或过小时，会影响预测的准确率，我用的是max_depth=15，min_samples_leaf叶子节点最少样本数，我用的是20，splitter: 特征划分标准，默认值为‘best’，在特征的所有划分点中找出最优的划分点，‘random’随机的在部分划分点中找局部最优的划分点。默认的‘best’适合样本量不大的时候，而如果样本数据量非常大，此时决策树构建推荐‘random’。数据集应该是数据量很大，但是我在用‘random’时，准确率并没有‘best’准确率高。
结果：

随机森林：首先先调既不会增加模型复杂度，又对模型影响最大的参数n_estimators（学习曲线），这里我取的是n_estimators=400，max_depth沿用决策树中15，其余值默认。
结果：

梯度提升决策树：n_estimators=400，max_depth=15。
结果综合：

代码：

代码：# -*- coding: utf-8 -*-
import numpy as np
from sklearn import datasets
from sklearn.metrics import accuracy_score
from sklearn.metrics import precision_recall_fscore_support as score
from scipy import stats
import matplotlib.pyplot as plot 
from sklearn.ensemble import AdaBoostClassifier
from sklearn.tree import DecisionTreeClassifier 
from sklearn.metrics import confusion_matrix
from sklearn.metrics import precision_score
from sklearn.metrics import recall_score
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import make_scorer, accuracy_score
from sklearn.model_selection import GridSearchCV  
X_mode = [None,None,None,None,None,None,None,None,None,None,None,None,None]; 
def workClassValue(x):   
return workClassValueHelper(str(x).replace("'", "").replace(" ", "").replace("b", "",1).replace(".", ""))  
def workClassValueHelper(xString):    
return {        
#workclass column values        
"Private": computeWeightWorkClass(1.0),     
"Self-emp-not-inc": computeWeightWorkClass(2.0),        
"Self-emp-inc": computeWeightWorkClass(3.0),        
"Federal-gov": computeWeightWorkClass(4.0),        
"Local-gov": computeWeightWorkClass(5.0),        
"State-gov": computeWeightWorkClass(6.0),        
"Without-pay": computeWeightWorkClass(7.0),        
"Never-worked": computeWeightWorkClass(8.0)    }.get(xString, getMode(1))  
def computeWeightWorkClass(x):    
return x  
def educationValue(x):    
return educationValueHelper(str(x).replace("'", "").replace(" ", "").replace("b", "",1).replace(".", ""))  
def educationValueHelper(xString):    
x = 2.0    
return {        
#education column values        
"Preschool": computeWeightEdu(1.0),        
"1st-4th": computeWeightEdu(2.0),        
"5th-6th": computeWeightEdu(3.0),       
"7th-8th": computeWeightEdu(4.0),        
"9th": computeWeightEdu(5.0),       
"10th": computeWeightEdu(6.0),        
"11th": computeWeightEdu(7.0),        
"12th": computeWeightEdu(8.0),        
"HS-grad": computeWeightEdu(9.0),        
"Some-college": computeWeightEdu(10.0),       
"Assoc-acdm": computeWeightEdu(11.0),        
"Assoc-voc": computeWeightEdu(12.0),       
"Prof-school": computeWeightEdu(13.0),        
"Bachelors": computeWeightEdu(14.0),        
"Masters": computeWeightEdu(15.0),        
"Doctorate": computeWeightEdu(16.0),    }.get(xString, getMode(2))  
def computeWeightEdu(x):    
return x  
def maritalStatusValue(x):    
return maritalStatusValueHelper(str(x).replace("'", "").replace(" ", "").replace("b", "", 1).replace(".", ""))  
def maritalStatusValueHelper(xString):    
return {        
#marital status column values        
"Never-married": 1.0,        
"Divorced": 2.0,        
"Separated": 3.0,        
"Widowed": 4.0,        
"Married-spouse-absent": 5.0,        
"Married-AF-spouse": 6.0,        
"Married-civ-spouse": 7.0    }.get(xString, getMode(3))  
def occupationValue(x):    
return occupationValueHelper(str(x).replace("'", "").replace(" ", "").replace("b", "",1).replace(".", ""))  
def occupationValueHelper(xString):    
return {        
#occupation column values        
"Tech-support": 1.0,        
"Craft-repair": 2.0,        
"Other-service": 3.0,        
"Sales": 4.0,        
"Exec-managerial": 5.0,        
"Prof-specialty": 6.0,        
"Handlers-cleaners": 7.0,        
"Machine-op-inspct": 8.0,        
"Adm-clerical": 9.0,        
"Farming-fishing": 10.0,        
"Transport-moving": 11.0,        
"Priv-house-serv": 12.0,        
"Protective-serv": 13.0,        
"Armed-Forces": 14.0    }.get(xString, getMode(4))  
def relationshipValue(x):    
return relationshipValueHelper(str(x).replace("'", "").replace(" ", "").replace("b", "",1).replace(".", ""))  
def relationshipValueHelper(xString):    
return {       
#relationship column values        
"Wife": 1.0,        
"Own-child": 2.0,        
"Husband": 3.0,        
"Not-in-family": 4.0,        
"Other-relative": 5.0,        
"Unmarried": 6.0    }.get(xString, getMode(5))  
def raceValue(x):    
return raceValueHelper(str(x).replace("'", "").replace(" ", "").replace("b", "",1).replace(".", ""))  
def raceValueHelper(xString):    
return {        
#race column values        
"White": 1.0,        
"Asian-Pac-Islander": 2.0,        
"Amer-Indian-Eskimo": 3.0,        
"Other": 4.0,        
"Black": 5.0,    }.get(xString, getMode(6))  
def genderValue(x):    
return genderValueHelper(str(x).replace("'", "").replace(" ", "").replace("b", "",1).replace(".", ""))  
def genderValueHelper(xString):    
return {        
#gender        
"Male": 1.0,        
"Female": 2.0    }.get(xString, getMode(7))  
def countryValue(x):    
return countryValueHelper(str(x).replace("'", "").replace(" ", "").replace("b", "",1).replace(".", ""))  
def countryValueHelper(xString):    
#print(xString + "1")    
return {        
#native country        
"United-States": 1.0,        
"Cambodia": 2.0,        
"England": 3.0,        
"Puerto-Rico": 4.0,        
"Canada": 5.0,        
"Germany": 6.0,        
"Outlying-US(Guam-USVI-etc)": 7.0,        
"India": 8.0,        
"Japan": 9.0,        
"Greece": 10.0,        
"South": 11.0,        
"China": 12.0,        
"Cuba": 13.0,        
"Iran": 14.0,       
"Honduras": 15.0,        
"Philippines": 16.0,        
"Italy": 17.0,        
"Poland": 18.0,        
"Jamaica": 19.0,        
"Vietnam": 20.0,        
"Mexico": 21.0,        
"Portugal": 22.0,        
"Ireland": 23.0,        
"France": 24.0,        
"Dominican-Republic": 25.0,        
"Laos": 26.0,        
"Ecuador": 27.0,        
"Taiwan": 28.0,        
"Haiti": 29.0,        
"Columbia": 30.0,        
"Hungary": 31.0,        
"Guatemala": 32.0,        
"Nicaragua": 33.0,        
"Scotland": 34.0,        
"Thailand": 35.0,        
"Yugoslavia": 36.0,        
"El-Salvador": 37.0,        
"Trinadad&Tobago": 38.0,        
"Peru": 39.0,        
"Hong": 40.0,        
"Holand-Netherlands": 41.0    }.get(xString, getMode(11))  
def salaryValue(x):    
return salaryValueHelper(str(x).replace("'", "").replace(" ", "").replace("b", "", 1).replace(".", ""))  
def salaryValueHelper(xString):    
return {        
#salary        
"<=50K": 1.0,        
">50K": -1.0    }.get(xString, getMode(12))  
def printValBefore(x):    
print(x)    return getMode(11)  def getMode(column):    
global X_mode    x = 0.0    if X_mode[0] is None:        
return x    
x = X_mode[column]    x = x    
#print(" new mode val " + str(x))    
return x  
def getModeArray(X_data):    
X_data_new = trimData(X_data)    
print(X_data_new.shape)     
Xmode = stats.mode(X_data_new)    
x = Xmode[0][0]    
print(x)     
return x 
def trimData(X_data):    
#get columns 1-2    X_data1 = X_data[:, :2]    
#print(X_data1.shape)     
#get column 4    X_data2 = np.expand_dims(X_data[:, 3], axis=1)    
#print(X_data2.shape)     
#get columns 6-14    X_data3 = X_data[:, 5:15]   
#print(X_data3.shape)     
#merge all sub X_data sets    
X_data_new = np.concatenate((np.concatenate((X_data1, X_data2), axis=1), X_data3), axis=1)    
print(X_data_new.shape)     
return X_data_new 
main functionif __name__ == '__main__':    
cancer_data = np.genfromtxt(fname='adult.csv',dtype='float', delimiter=',',missing_values="nan", filling_values=1, converters={
1: workClassValue,         
3: educationValue, 
5: maritalStatusValue, 
6: occupationValue, 
7: relationshipValue, 
8: raceValue, 
9: genderValue, 
13: countryValue, 
14: salaryValue})    
adata = np.genfromtxt(fname='adult.csv', dtype='float', delimiter=',', converters= {
1: workClassValue, 
3: educationValue,                   
5: maritalStatusValue, 
6: occupationValue, 
7: relationshipValue, 
8: raceValue, 
9: genderValue, 
13: countryValue, 
14: salaryValue})            
adata_test = np.genfromtxt(
fname='adult.test.csv', dtype='float', delimiter=',', converters= {
1: workClassValue, 
3: educationValue,                    
5: maritalStatusValue, 
6: occupationValue, 
7: relationshipValue, 
8: raceValue, 
9: genderValue, 
13: countryValue, 
14: salaryValue})   
#removing columns 3 and 5    
adata = trimData(adata)    
adata_y = adata[:,-1];    
adata = adata[:,0:14]    
#removing columns 3 and 5    
adata_test = trimData(adata_test)    
adata_test_y = adata_test[:,-1];    
adata_test = adata_test[:,0:14];    
#X_data is your training Y_data is your testing    
clf = RandomForestClassifier(oob_score=False, random_state=20, criterion='entropy', max_depth=15,n_estimators=400)     
clf = DecisionTreeClassifier(criterion='entropy', splitter='best',min_samples_leaf=20,max_depth=15)     clf.fit(adata, adata_y)    
bdt = AdaBoostClassifier(DecisionTreeClassifier(max_depth=5),algorithm="SAMME",n_estimators=400)    bdt.fit(adata, adata_y)     
ypred = clf.predict(adata_test);    
ypred = bdt.predict(adata_test);    
p,r,f,s = score(adata_test_y,ypred)    
print('F-scrore: ', f, " Combined: ",(f[0]+f[1])/2)        
cnf_matrix = confusion_matrix(adata_test_y, ypred)    
precision = precision_score(adata_test_y,ypred)    
recall = recall_score(adata_test_y, ypred)    
F_score = 2*((precision*recall)/(precision+recall))        
print("Accuracy ", accuracy_score(adata_test_y,ypred))    
print("Confusion Matrix:\n", cnf_matrix)    
print("Precision: ", precision*100)    
print("Recall: ", recall*100)    
print("F-score: ", F_score)