choice model using xgboost
Published:
This is a blog post that delves into the application of the XGBoost algorithm, a popular machine learning technique, in choice modeling. The post provides a comprehensive guide on how to implement this algorithm, discussing its advantages and potential challenges. It serves as a valuable resource for anyone interested in enhancing their machine learning skills, particularly in the field of choice modeling.
STEP1 数据预处理和分析
1.数据预处理(扩样+划分训练集和验证集)
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.preprocessing import OneHotEncoder, StandardScaler
from sklearn.compose import ColumnTransformer
from sklearn.model_selection import train_test_split
#导入数据与所有初步的数据处理都保存在原始表df0，此后所有的分析都是从这个表中抽取数据
df0 = pd.read_csv(r'C:\Users\HONOR\Desktop\统计建模比赛\csv\tripv2pub.csv')
#我们初步对出行方式做了分类，TRPTRANS值落入“pubgroup”数组的是小组认为的公共交通，ISPUB值为1；其他小组认为并非公共交通，ISPUB值为0
pubgroup=[8,9,11,12,15,16,17,18,20,22]
df0['ISPUB']=df0['TRPTRANS'].apply(lambda x:1 if x in pubgroup else 0)
df0['AVPUB']=1
df0['AVXPUB']=1
def clean_data(df1):
df1 = df1[df1['DRIVER'] != -1]
df1 = df1[df1['TRIPPURP'] != -9]
df1 = df1[df1['TRPMILES'] != -9]
df1 = df1[df1['TDWKND'] != -9]
df1 = df1[df1['WORKER'] != -1]
return df1
df0 = clean_data(df0)
# 挑选出所有所需的列名，保存至组“selected_columns”中
selected_columns = ['ISPUB','PUBTRANS','WTTRDFIN','WTTRDFIN5D','WTTRDFIN2D','WORKER', 'WALK', 'URBRUR', 'TRPMILES', 'TRIPPURP', 'TDWKND', 'R_AGE', 'DRIVER','HHFAMINC_IMP','AVPUB','AVXPUB',
'NUMONTRP','GASPRICE','RAIL','DRVRCNT','HHVEHCNT','HHSIZE']
# 调用函数clean_df_totest清理原始表df0
df0 = clean_data(df0)
df0 = df0[selected_columns]
df0['PUBTRANS']=df0['PUBTRANS'].apply(lambda x:1 if x==1 else 0)
df0['RAIL']=df0['RAIL'].apply(lambda x:1 if x==1 else 0)
df0['DRIVER']=df0['DRIVER'].apply(lambda x:1 if x==1 else 0)
df0['WORKER']=df0['WORKER'].apply(lambda x:1 if x==1 else 0)
df0['URBRUR']=df0['URBRUR'].apply(lambda x:1 if x==1 else 0)
df0['TDWKND']=df0['TDWKND'].apply(lambda x:1 if x==1 else 0)
df0['frequency']=(1.20878e-07 * df0['WTTRDFIN']).astype(int)
df0['frequency'] = df0['frequency'].apply(lambda x: 1 if x < 1 else x)
df0
# 扩样
df_replicated = df0.loc[np.repeat(df0.index.values, df0['frequency'])]
df_rep0=df_replicated
df_rep0.to_csv('data_whole.csv', index=False) # 洗过、扩样过的，还未分训练集和测试集的数据
# 分类变量和连续变量
categorical_features = ['WORKER', 'URBRUR', 'TRIPPURP', 'TDWKND', 'DRIVER', 'RAIL', 'AVPUB', 'AVXPUB']
numeric_features = ['WALK', 'TRPMILES', 'HHFAMINC_IMP', 'NUMONTRP', 'GASPRICE', 'DRVRCNT', 'HHVEHCNT', 'HHSIZE','R_AGE']
# 划分训练集和测试集
X_train, X_test, y_train, y_test = train_test_split(df_rep0[categorical_features + numeric_features], df_rep0['PUBTRANS'].values, test_size=0.2, random_state=42, shuffle=True)
# 将y_train添加到X_train df_totestFrame中作为一列
X_train['PUBTRANS'] = y_train
X_test['PUBTRANS'] = y_test
# 输出CSV文件
X_train.to_csv('data_train.csv', index=False)
X_test.to_csv('data_test.csv', index=False)
2. 数据探索
# 变量关联发掘(对于df1)
categorical_features = ['WORKER', 'URBRUR', 'TRIPPURP', 'TDWKND', 'DRIVER', 'RAIL']
numeric_features = ['WALK', 'TRPMILES', 'HHFAMINC_IMP', 'NUMONTRP', 'GASPRICE', 'DRVRCNT', 'HHVEHCNT', 'HHSIZE','R_AGE']
y_feature=['PUBTRANS']
selected_columns=numeric_features+categorical_features+y_feature
df1 = df_rep0[selected_columns]
#由于数据集包含了分类变量和连续变量，所以我们可以使用Kendall's Tau相关系数来计算相关性。
# 计算Kendall's Tau相关系数
corr = df1.corr(method='kendall')
# 创建一个热力图
plt.figure(figsize=(12, 10))
sns.heatmap(corr, annot=True, fmt=".2f", cmap='coolwarm', cbar=True, square=True)
plt.xticks(rotation=45) # 旋转x轴标签
plt.yticks(rotation=45) # 旋转y轴标签
plt.show()
STEP2 离散选择模型-logit & probit
0. 数据导入
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import biogeme.biogeme as bio
import biogeme.results as res
from biogeme import models
import biogeme.database as db
import biogeme.version as ver
from biogeme.expressions import Variable
from biogeme.expressions import Beta
import seaborn as sns
from sklearn.metrics import accuracy_score, recall_score, precision_score, f1_score, roc_auc_score, confusion_matrix
from biogeme.expressions import bioNormalCdf, log, Elem
def build(df0, positive_features, negative_features, model_type, model_name):
# 建立数据库
database = db.Database('data', df0)
globals().update(database.variables)
ASC = Beta('ASC', 0, None, None, 0)
# positive features
positive_betas = {f"B_{feature}": Beta(f"B_{feature}", 0, None, None, 0) for feature in positive_features}
# negative features
negative_betas = {f"B_{feature}": Beta(f"B_{feature}", 0, None, None, 0) for feature in negative_features}
V1 = ASC + sum([positive_betas[f"B_{feature}"]*globals()[feature] for feature in positive_features])
V2 = sum([negative_betas[f"B_{feature}"]*globals()[feature] for feature in negative_features])
av = {1: AVPUB, 0: AVXPUB}
V = {1: V1, 0: V2}
if model_type == 'logit':
logprob = models.loglogit(V, av, PUBTRANS)
elif model_type == 'probit':
P = {1: bioNormalCdf(V1-V2),
0: bioNormalCdf(V2-V1)}
logprob = log(Elem(P, PUBTRANS))
else:
raise ValueError("Invalid model type. Choose either 'logit' or 'probit'.")
biogeme = bio.BIOGEME(database, logprob)
biogeme.modelName = model_name
results = biogeme.estimate()
print(results)
return results
def calculate_metrics(y_test, y_pred):
accuracy = accuracy_score(y_test, y_pred)
recall = recall_score(y_test, y_pred)
precision = precision_score(y_test, y_pred)
f1 = f1_score(y_test, y_pred)
roc_auc = roc_auc_score(y_test, y_pred)
cm = confusion_matrix(y_test, y_pred)
print(f"Accuracy: {accuracy}")
print(f"Recall: {recall}")
print(f"Precision: {precision}")
print(f"F1 Score: {f1}")
print(f"ROC AUC Score: {roc_auc}")
# 绘制混淆矩阵的热力图
plt.figure(figsize=(10, 7))
sns.heatmap(cm, annot=True, fmt='d', cmap='Blues')
plt.xlabel('Predicted')
plt.ylabel('True')
plt.show()
def softmax(x):
e_x = np.exp(x-np.max(x))
return e_x / e_x.sum(axis=0)
def eval(beta_values, positive_features, negative_features,df_totest):
# 从模型结果中获取参数值
ASC = beta_values['ASC']
positive_betas = {f"B_{feature}": beta_values[f"B_{feature}"] for feature in positive_features}
negative_betas = {f"B_{feature}": beta_values[f"B_{feature}"] for feature in negative_features}
# 计算V1和V2
df_totest['V1'] = ASC + sum([positive_betas[f"B_{feature}"] * df_totest[feature] for feature in positive_features])
df_totest['V2'] = sum([negative_betas[f"B_{feature}"] * df_totest[feature] for feature in negative_features])
# 计算概率
df_totest[['p1', 'p2']] = df_totest.apply(lambda row: softmax([row['V1'], row['V2']]), axis=1, result_type='expand')
# 进行选择
df_totest['choice'] = np.where(df_totest['p1'] > df_totest['p2'], 1, 0)
# 评估
return calculate_metrics(df_totest['PUBTRANS'], df_totest['choice'])
# 全部的基本features（没有经过变换）
#positive_features = ['TRPMILES', 'GASPRICE', 'URBRUR', 'RAIL']
#negative_features = ['WORKER','TRIPPURP','TDWKND','DRIVER','WALK','HHFAMINC_IMP',
# 'NUMONTRP', 'DRVRCNT', 'HHVEHCNT', 'HHSIZE','R_AGE']
1. 基础模型
#探索性 logit
#（1）选指标
#（如果要做指标变换，请对训练集和测试集都进行变换！！，并将其新的列名加入positive_features和negative_features）
positive_features = ['TRPMILES', 'GASPRICE', 'URBRUR', 'TRIPPURP','RAIL']
negative_features = ['WALK','HHFAMINC_IMP','NUMONTRP','DRVRCNT','HHVEHCNT','HHSIZE','R_AGE','WORKER','TDWKND','DRIVER']
#（2）读训练集
df0 = pd.read_csv('data_train.csv')
#（3）建模结果
results=build(df0, positive_features, negative_features, 'logit','logit-discover')
#（4）获取估计参数
parameters=results.getEstimatedParameters()
print(parameters)
beta_values=parameters['Value']
parameters.to_csv('logit-discover-parameters.csv')# 保存参数
#（5）评测
df1=pd.read_csv('data_test.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test1.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test2.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test3.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test4.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test5.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
File biogeme.toml has been created
Results for model logit-discover
Output file (HTML): logit-discover.html
Nbr of parameters: 16
Sample size: 27448
Excluded data: 0
Init log likelihood: -19025.5
Final log likelihood: -1166.541
Likelihood ratio test (init): 35717.93
Rho square (init): 0.939
Rho bar square (init): 0.938
Akaike Information Criterion: 2365.082
Bayesian Information Criterion: 2496.603
Final gradient norm: 0.05410666
ASC : -4.17[0.544 -7.67 1.67e-14][0.522 -7.99 1.33e-15]
B_DRIVER : 1.66[0.177 9.37 0][0.175 9.48 0]
B_DRVRCNT : -0.615[0.114 -5.39 6.87e-08][0.114 -5.38 7.5e-08]
B_GASPRICE : 0.00482[0.000889 5.42 5.85e-08][0.0008 6.03 1.68e-09]
B_HHFAMINC_IMP : 0.106[0.0237 4.45 8.62e-06][0.0254 4.16 3.17e-05]
B_HHSIZE : 0.094[0.0485 1.94 0.0524][0.0502 1.87 0.0609]
B_HHVEHCNT : 1.6[0.108 14.9 0][0.116 13.8 0]
B_NUMONTRP : -0.0172[0.00876 -1.96 0.0502][0.0106 -1.62 0.106]
B_RAIL : 1.35[0.133 10.1 0][0.131 10.3 0]
B_R_AGE : 0.00273[0.00389 0.702 0.483][0.00398 0.686 0.493]
B_TDWKND : 1.26[0.174 7.25 4.04e-13][0.188 6.73 1.74e-11]
B_TRIPPURP : -0.0105[0.0418 -0.252 0.801][0.0438 -0.241 0.81]
B_TRPMILES : 0.000309[0.000812 0.38 0.704][0.000303 1.02 0.308]
B_URBRUR : 0.174[0.279 0.622 0.534][0.292 0.595 0.552]
B_WALK : 0.189[0.013 14.5 0][0.0119 15.8 0]
B_WORKER : -0.693[0.145 -4.8 1.62e-06][0.155 -4.47 7.81e-06]
('B_DRIVER', 'ASC'): -0.00393 -0.0408 10.1 0 -0.00397 -0.0434 10.5 0
('B_DRVRCNT', 'ASC'): 0.0055 0.0887 6.52 7.02e-11 0.00654 0.109 6.81 9.75e-12
('B_DRVRCNT', 'B_DRIVER'): -0.0125 -0.619 -8.63 0 -0.0112 -0.561 -8.84 0
('B_GASPRICE', 'ASC'): -0.000329 -0.681 7.67 1.67e-14 -0.000265 -0.633 7.99 1.33e-15
('B_GASPRICE', 'B_DRIVER'): 1.57e-05 0.0996 -9.35 0 1.22e-05 0.0869 -9.45 0
('B_GASPRICE', 'B_DRVRCNT'): -9.48e-06 -0.0934 5.43 5.55e-08 -1.09e-06 -0.012 5.42 5.95e-08
('B_HHFAMINC_IMP', 'ASC'): 0.000139 0.0108 7.86 3.77e-15 -0.00166 -0.125 8.13 4.44e-16
('B_HHFAMINC_IMP', 'B_DRIVER'): 0.000242 0.0576 -8.76 0 0.00063 0.142 -8.96 0
('B_HHFAMINC_IMP', 'B_DRVRCNT'): -0.000502 -0.185 5.97 2.36e-09 -0.000558 -0.192 5.92 3.3e-09
('B_HHFAMINC_IMP', 'B_GASPRICE'): 1.06e-06 0.0505 4.25 2.13e-05 9.23e-07 0.0455 3.97 7.06e-05
('B_HHSIZE', 'ASC'): 0.00642 0.244 7.99 1.33e-15 0.00678 0.259 8.34 0
('B_HHSIZE', 'B_DRIVER'): 0.00332 0.386 -9.51 0 0.00326 0.371 -9.59 0
('B_HHSIZE', 'B_DRVRCNT'): -0.00256 -0.463 4.96 7.17e-07 -0.00225 -0.392 5 5.66e-07
('B_HHSIZE', 'B_GASPRICE'): -1.82e-06 -0.0421 1.84 0.0659 -2.65e-06 -0.0659 1.78 0.0757
('B_HHSIZE', 'B_HHFAMINC_IMP'): 0.000158 0.138 -0.225 0.822 0.000254 0.2 -0.223 0.824
('B_HHVEHCNT', 'ASC'): -0.0058 -0.0991 10.2 0 -0.00641 -0.106 10.6 0
('B_HHVEHCNT', 'B_DRIVER'): 0.00215 0.112 -0.296 0.767 -8.8e-05 -0.00433 -0.277 0.782
('B_HHVEHCNT', 'B_DRVRCNT'): -0.0066 -0.537 11.4 0 -0.00583 -0.44 11.3 0
('B_HHVEHCNT', 'B_GASPRICE'): 1.16e-05 0.121 14.8 0 9.61e-06 0.104 13.8 0
('B_HHVEHCNT', 'B_HHFAMINC_IMP'): -0.000514 -0.201 13 0 -0.000659 -0.224 12.1 0
('B_HHVEHCNT', 'B_HHSIZE'): -4.02e-05 -0.00769 12.7 0 -0.00102 -0.176 11.2 0
('B_NUMONTRP', 'ASC'): 0.000338 0.071 7.65 2e-14 0.000395 0.0714 7.97 1.55e-15
('B_NUMONTRP', 'B_DRIVER'): 6.65e-06 0.00429 -9.46 0 7.28e-05 0.0392 -9.58 0
('B_NUMONTRP', 'B_DRVRCNT'): -6.71e-05 -0.0671 5.2 1.97e-07 -0.000166 -0.137 5.14 2.72e-07
('B_NUMONTRP', 'B_GASPRICE'): -5.6e-07 -0.0718 -2.48 0.0132 -5.27e-07 -0.062 -2.06 0.0398
('B_NUMONTRP', 'B_HHFAMINC_IMP'): -1.08e-05 -0.0518 -4.77 1.82e-06 -1.48e-05 -0.055 -4.38 1.2e-05
('B_NUMONTRP', 'B_HHSIZE'): 1.01e-05 0.0238 -2.27 0.0234 1.82e-05 0.0342 -2.18 0.029
('B_NUMONTRP', 'B_HHVEHCNT'): 4.89e-05 0.0518 -15 0 0.000102 0.0828 -14 0
('B_RAIL', 'ASC'): 0.00289 0.0399 9.96 0 0.00599 0.0873 10.5 0
('B_RAIL', 'B_DRIVER'): -0.000265 -0.0112 -1.39 0.166 -0.000409 -0.0177 -1.4 0.162
('B_RAIL', 'B_DRVRCNT'): -0.000705 -0.0464 11 0 -0.000476 -0.0317 11.1 0
('B_RAIL', 'B_GASPRICE'): -4.57e-06 -0.0385 10.1 0 -1.52e-05 -0.144 10.2 0
('B_RAIL', 'B_HHFAMINC_IMP'): 0.00119 0.375 9.86 0 0.00108 0.324 9.92 0
('B_RAIL', 'B_HHSIZE'): -3.35e-05 -0.00518 8.85 0 0.000531 0.0805 9.18 0
('B_RAIL', 'B_HHVEHCNT'): -0.000843 -0.0587 -1.42 0.155 -0.0013 -0.0854 -1.37 0.17
('B_RAIL', 'B_NUMONTRP'): 3.77e-05 0.0323 10.3 0 3.48e-06 0.0025 10.4 0
('B_R_AGE', 'ASC'): 0.000928 0.439 7.7 1.33e-14 0.00104 0.501 8.02 1.11e-15
('B_R_AGE', 'B_DRIVER'): -0.000159 -0.23 -9.3 0 -0.000162 -0.232 -9.41 0
('B_R_AGE', 'B_DRVRCNT'): 8.17e-05 0.184 5.45 5.05e-08 0.000105 0.232 5.44 5.23e-08
('B_R_AGE', 'B_GASPRICE'): -2.75e-07 -0.0795 -0.516 0.606 -1.91e-07 -0.0601 -0.51 0.61
('B_R_AGE', 'B_HHFAMINC_IMP'): -8.07e-06 -0.0875 -4.22 2.46e-05 -1.94e-05 -0.193 -3.89 9.99e-05
('B_R_AGE', 'B_HHSIZE'): 3.08e-05 0.164 -1.9 0.0571 4.77e-05 0.239 -1.85 0.0644
('B_R_AGE', 'B_HHVEHCNT'): -5.44e-05 -0.13 -14.8 0 -6.34e-05 -0.137 -13.7 0
('B_R_AGE', 'B_NUMONTRP'): 3.16e-07 0.00928 2.08 0.0374 5.6e-07 0.0133 1.76 0.078
('B_R_AGE', 'B_RAIL'): -8.02e-06 -0.0155 -10.1 0 -8.73e-06 -0.0167 -10.2 0
('B_TDWKND', 'ASC'): -0.0025 -0.0264 9.45 0 -0.0133 -0.136 9.4 0
('B_TDWKND', 'B_DRIVER'): 0.0022 0.0711 -1.66 0.0977 0.00349 0.106 -1.63 0.103
('B_TDWKND', 'B_DRVRCNT'): -0.00236 -0.119 8.57 0 -0.00461 -0.215 7.83 4.88e-15
('B_TDWKND', 'B_GASPRICE'): 1.32e-05 0.0852 7.23 4.85e-13 2.28e-05 0.152 6.7 2.02e-11
('B_TDWKND', 'B_HHFAMINC_IMP'): -0.000132 -0.032 6.56 5.39e-11 0.000353 0.0741 6.17 6.82e-10
('B_TDWKND', 'B_HHSIZE'): 0.000982 0.116 6.67 2.52e-11 0.00176 0.186 6.32 2.7e-10
('B_TDWKND', 'B_HHVEHCNT'): 0.00256 0.136 -1.76 0.0779 0.00307 0.141 -1.64 0.101
('B_TDWKND', 'B_NUMONTRP'): 4.5e-05 0.0295 7.35 1.92e-13 5.34e-05 0.0268 6.82 9.29e-12
('B_TDWKND', 'B_RAIL'): -0.000423 -0.0182 -0.396 0.692 0.00195 0.0791 -0.397 0.691
('B_TDWKND', 'B_R_AGE'): -1.03e-06 -0.00152 7.24 4.61e-13 -7.98e-06 -0.0107 6.71 1.96e-11
('B_TRIPPURP', 'ASC'): -0.00562 -0.248 7.49 6.79e-14 -0.00576 -0.252 7.78 7.33e-15
('B_TRIPPURP', 'B_DRIVER'): 0.000696 0.0941 -9.38 0 0.00118 0.154 -9.61 0
('B_TRIPPURP', 'B_DRVRCNT'): -0.000598 -0.126 4.79 1.68e-06 -0.00136 -0.272 4.54 5.56e-06
('B_TRIPPURP', 'B_GASPRICE'): -1.44e-06 -0.0389 -0.368 0.713 -1.17e-06 -0.0335 -0.351 0.726
('B_TRIPPURP', 'B_HHFAMINC_IMP'): 3.9e-05 0.0393 -2.46 0.0139 0.00015 0.135 -2.44 0.0146
('B_TRIPPURP', 'B_HHSIZE'): 0.000176 0.0867 -1.71 0.0873 0.000342 0.156 -1.71 0.0876
('B_TRIPPURP', 'B_HHVEHCNT'): 0.000366 0.0814 -14.4 0 0.000106 0.0209 -13.1 0
('B_TRIPPURP', 'B_NUMONTRP'): 2.78e-05 0.0759 0.157 0.875 4.65e-05 0.1 0.15 0.881
('B_TRIPPURP', 'B_RAIL'): -8.1e-05 -0.0146 -9.71 0 -0.000635 -0.11 -9.52 0
('B_TRIPPURP', 'B_R_AGE'): -1.43e-05 -0.0883 -0.314 0.754 -2.49e-05 -0.143 -0.298 0.766
('B_TRIPPURP', 'B_TDWKND'): 0.000169 0.0232 -7.15 8.62e-13 0.000402 0.0489 -6.68 2.42e-11
('B_TRPMILES', 'ASC'): 1.02e-05 0.0231 7.67 1.67e-14 -1.4e-06 -0.00886 7.99 1.33e-15
('B_TRPMILES', 'B_DRIVER'): 2.68e-06 0.0187 -9.37 0 7.04e-07 0.0133 -9.47 0
('B_TRPMILES', 'B_DRVRCNT'): -3.19e-06 -0.0345 5.4 6.82e-08 -2.96e-06 -0.0854 5.38 7.44e-08
('B_TRPMILES', 'B_GASPRICE'): -1.93e-08 -0.0268 -3.7 0.000215 -3e-09 -0.0124 -5.25 1.49e-07
('B_TRPMILES', 'B_HHFAMINC_IMP'): -1.87e-07 -0.00971 -4.43 9.33e-06 -2.46e-07 -0.032 -4.15 3.38e-05
('B_TRPMILES', 'B_HHSIZE'): -3.63e-07 -0.00921 -1.93 0.0532 -6.21e-07 -0.0409 -1.87 0.0618
('B_TRPMILES', 'B_HHVEHCNT'): 7.05e-06 0.0806 -14.9 0 5.69e-06 0.162 -13.8 0
('B_TRPMILES', 'B_NUMONTRP'): 1.13e-06 0.159 2.01 0.044 6.12e-07 0.19 1.65 0.098
('B_TRPMILES', 'B_RAIL'): 1.72e-06 0.0159 -10.1 0 -1.26e-07 -0.00316 -10.3 0
('B_TRPMILES', 'B_R_AGE'): 2.38e-08 0.00753 -0.611 0.541 5.53e-09 0.00458 -0.607 0.544
('B_TRPMILES', 'B_TDWKND'): 1.24e-06 0.00878 -7.25 4.09e-13 1.23e-06 0.0217 -6.72 1.76e-11
('B_TRPMILES', 'B_TRIPPURP'): -1.77e-06 -0.0522 0.26 0.795 -1.95e-06 -0.147 0.248 0.804
('B_URBRUR', 'ASC'): -0.0677 -0.446 6.09 1.11e-09 -0.0731 -0.479 6.12 9.27e-10
('B_URBRUR', 'B_DRIVER'): -0.000898 -0.0182 -4.46 8.26e-06 0.000965 0.0189 -4.4 1.06e-05
('B_URBRUR', 'B_DRVRCNT'): -5.79e-05 -0.00182 2.61 0.00895 -0.000343 -0.0103 2.51 0.0121
('B_URBRUR', 'B_GASPRICE'): -6.06e-06 -0.0244 0.604 0.546 9.66e-06 0.0413 0.578 0.563
('B_URBRUR', 'B_HHFAMINC_IMP'): 0.000109 0.0165 0.243 0.808 0.00177 0.239 0.237 0.812
('B_URBRUR', 'B_HHSIZE'): 0.000622 0.046 0.283 0.777 0.00135 0.092 0.273 0.785
('B_URBRUR', 'B_HHVEHCNT'): -0.00214 -0.071 -4.66 3.14e-06 -0.000241 -0.00713 -4.54 5.7e-06
('B_URBRUR', 'B_NUMONTRP'): -0.000109 -0.0446 0.682 0.495 -0.000118 -0.0381 0.652 0.514
('B_URBRUR', 'B_RAIL'): -0.00791 -0.213 -3.53 0.000422 -0.00303 -0.079 -3.57 0.000351
('B_URBRUR', 'B_R_AGE'): -8.08e-06 -0.00744 0.612 0.541 -5.79e-05 -0.0498 0.585 0.559
('B_URBRUR', 'B_TDWKND'): 0.00116 0.0239 -3.35 0.000814 0.00901 0.164 -3.41 0.000661
('B_URBRUR', 'B_TRIPPURP'): 0.000464 0.0398 0.656 0.512 -0.000581 -0.0455 0.62 0.535
('B_URBRUR', 'B_TRPMILES'): 5.79e-06 0.0255 0.62 0.535 1.45e-05 0.164 0.594 0.553
('B_WALK', 'ASC'): -0.000921 -0.13 7.99 1.33e-15 -0.000847 -0.136 8.32 0
('B_WALK', 'B_DRIVER'): -0.000212 -0.0918 -8.23 2.22e-16 -0.000327 -0.157 -8.29 0
('B_WALK', 'B_DRVRCNT'): 0.000107 0.0722 7.06 1.67e-12 0.000133 0.0977 7.06 1.67e-12
('B_WALK', 'B_GASPRICE'): 5.97e-07 0.0516 14.1 0 3.82e-07 0.0401 15.4 0
('B_WALK', 'B_HHFAMINC_IMP'): 3.51e-05 0.114 3.23 0.00125 5.22e-05 0.173 3.18 0.00146
('B_WALK', 'B_HHSIZE'): -1.17e-05 -0.0185 1.87 0.061 4.35e-05 0.0728 1.86 0.0625
('B_WALK', 'B_HHVEHCNT'): -0.00023 -0.164 -12.8 0 -0.000514 -0.372 -11.7 0
('B_WALK', 'B_NUMONTRP'): 3.49e-06 0.0306 13.3 0 3.77e-06 0.0299 13.1 0
('B_WALK', 'B_RAIL'): 8.62e-05 0.0497 -8.72 0 0.000134 0.0858 -8.88 0
('B_WALK', 'B_R_AGE'): -3.54e-06 -0.07 13.4 0 -5.35e-06 -0.113 14.3 0
('B_WALK', 'B_TDWKND'): 8.91e-06 0.00393 -6.16 7.43e-10 -0.000107 -0.0477 -5.69 1.24e-08
('B_WALK', 'B_TRIPPURP'): -7.84e-05 -0.144 4.38 1.21e-05 2.36e-05 0.0453 4.44 9.06e-06
('B_WALK', 'B_TRPMILES'): -2.23e-07 -0.0212 14.4 0 -4.78e-07 -0.133 15.8 0
('B_WALK', 'B_URBRUR'): -0.000174 -0.0478 0.0534 0.957 -0.00038 -0.11 0.0509 0.959
('B_WORKER', 'ASC'): 0.0191 0.242 6.59 4.29e-11 0.0286 0.353 7.11 1.18e-12
('B_WORKER', 'B_DRIVER'): -0.00746 -0.291 -9.08 0 -0.01 -0.368 -8.61 0
('B_WORKER', 'B_DRVRCNT'): 0.00332 0.201 -0.472 0.637 0.00364 0.205 -0.452 0.652
('B_WORKER', 'B_GASPRICE'): -5.14e-06 -0.0399 -4.83 1.38e-06 -7.64e-07 -0.00615 -4.5 6.76e-06
('B_WORKER', 'B_HHFAMINC_IMP'): -0.00107 -0.311 -5.2 1.98e-07 -0.00132 -0.336 -4.83 1.36e-06
('B_WORKER', 'B_HHSIZE'): -0.000413 -0.0589 -5.08 3.87e-07 8.01e-05 0.0103 -4.84 1.27e-06
('B_WORKER', 'B_HHVEHCNT'): -0.000644 -0.0414 -12.5 0 -0.000599 -0.0333 -11.7 0
('B_WORKER', 'B_NUMONTRP'): 3.99e-05 0.0315 -4.68 2.9e-06 0.000149 0.0905 -4.38 1.21e-05
('B_WORKER', 'B_RAIL'): -0.00115 -0.0596 -10.1 0 -0.000112 -0.00549 -10 0
('B_WORKER', 'B_R_AGE'): 0.000159 0.282 -4.85 1.23e-06 0.000222 0.359 -4.53 5.95e-06
('B_WORKER', 'B_TDWKND'): -0.00279 -0.111 -8.21 2.22e-16 -0.00434 -0.149 -7.5 6.24e-14
('B_WORKER', 'B_TRIPPURP'): -0.00142 -0.234 -4.28 1.88e-05 -0.00176 -0.258 -3.98 6.99e-05
('B_WORKER', 'B_TRPMILES'): 5.79e-06 0.0494 -4.8 1.59e-06 6.39e-06 0.136 -4.47 7.7e-06
('B_WORKER', 'B_URBRUR'): 0.000395 0.00979 -2.77 0.00564 -0.00771 -0.17 -2.46 0.0141
('B_WORKER', 'B_WALK'): -3.53e-05 -0.0188 -6.07 1.31e-09 -2.49e-05 -0.0135 -5.66 1.49e-08
Value Rob. Std err Rob. t-test Rob. p-value
ASC -4.172927 0.522371 -7.988430 1.332268e-15
B_DRIVER 1.660109 0.175189 9.476106 0.000000e+00
B_DRVRCNT -0.615397 0.114413 -5.378756 7.500238e-08
B_GASPRICE 0.004824 0.000800 6.026062 1.680027e-09
B_HHFAMINC_IMP 0.105510 0.025359 4.160582 3.174372e-05
B_HHSIZE 0.094037 0.050169 1.874406 6.087448e-02
B_HHVEHCNT 1.601801 0.115904 13.820105 0.000000e+00
B_NUMONTRP -0.017155 0.010609 -1.616977 1.058832e-01
B_RAIL 1.351137 0.131445 10.279118 0.000000e+00
B_R_AGE 0.002731 0.003980 0.686100 4.926500e-01
B_TDWKND 1.263520 0.187847 6.726324 1.740030e-11
B_TRIPPURP -0.010542 0.043781 -0.240785 8.097216e-01
B_TRPMILES 0.000309 0.000303 1.018485 3.084474e-01
B_URBRUR 0.173561 0.291851 0.594690 5.520504e-01
B_WALK 0.188509 0.011899 15.841981 0.000000e+00
B_WORKER -0.693443 0.155125 -4.470223 7.813823e-06
Accuracy: 0.9861576570013114
Recall: 0.16091954022988506
Precision: 0.3888888888888889
F1 Score: 0.22764227642276422
ROC AUC Score: 0.5788363934915659
Accuracy: 0.9218241042345277
Recall: 0.181360201511335
Precision: 0.8674698795180723
F1 Score: 0.3
ROC AUC Score: 0.5892702058569235
Accuracy: 0.8296675191815857
Recall: 0.17884130982367757
Precision: 0.9102564102564102
F1 Score: 0.29894736842105263
ROC AUC Score: 0.5871741850787194
Accuracy: 0.7513227513227513
Recall: 0.17884130982367757
Precision: 0.9594594594594594
F1 Score: 0.30148619957537154
ROC AUC Score: 0.5878007845014717
Accuracy: 0.6706948640483383
Recall: 0.17884130982367757
Precision: 0.9861111111111112
F1 Score: 0.302771855010661
ROC AUC Score: 0.5885817287373422
Accuracy: 0.5826972010178118
Recall: 0.17884130982367757
Precision: 0.9726027397260274
F1 Score: 0.3021276595744681
ROC AUC Score: 0.5868499608244352
#基础模型 logit
#（1）选指标
#（如果要做指标变换，请对训练集和测试集都进行变换！！，并将其新的列名加入positive_features和negative_features）
positive_features = ['GASPRICE', 'RAIL']
negative_features = ['WALK','HHFAMINC_IMP','DRVRCNT','HHVEHCNT','WORKER','TDWKND','DRIVER']
#（2）读训练集
df0 = pd.read_csv('data_train.csv')
#（3）建模（注意类型）
results=build(df0, positive_features, negative_features, 'logit','logit-0')
#（4）获取估计参数
parameters=results.getEstimatedParameters()
beta_values=parameters['Value']
print(parameters)
parameters.to_csv('logit-0-parameters.csv')# 保存估计参数
#（5）评测
df1=pd.read_csv('data_test.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test1.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test2.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test3.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test4.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test5.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
Results for model logit-0
Output file (HTML): logit-0.html
Nbr of parameters: 10
Sample size: 27448
Excluded data: 0
Init log likelihood: -19025.5
Final log likelihood: -1170.65
Likelihood ratio test (init): 35709.71
Rho square (init): 0.938
Rho bar square (init): 0.938
Akaike Information Criterion: 2361.299
Bayesian Information Criterion: 2443.5
Final gradient norm: 0.001319092
ASC : -4.38[0.393 -11.1 0][0.345 -12.7 0]
B_DRIVER : 1.55[0.154 10 0][0.151 10.3 0]
B_DRVRCNT : -0.536[0.0953 -5.63 1.81e-08][0.0946 -5.67 1.42e-08]
B_GASPRICE : 0.00483[0.000882 5.47 4.43e-08][0.000799 6.05 1.47e-09]
B_HHFAMINC_IMP : 0.0979[0.0233 4.2 2.62e-05][0.0243 4.03 5.69e-05]
B_HHVEHCNT : 1.62[0.106 15.3 0][0.114 14.2 0]
B_RAIL : 1.37[0.13 10.6 0][0.132 10.4 0]
B_TDWKND : 1.23[0.172 7.15 8.97e-13][0.183 6.74 1.6e-11]
B_WALK : 0.19[0.0128 14.9 0][0.0119 15.9 0]
B_WORKER : -0.695[0.134 -5.18 2.2e-07][0.141 -4.94 7.89e-07]
('B_DRIVER', 'ASC'): -0.00204 -0.0336 13.9 0 0.000263 0.00505 15.8 0
('B_DRVRCNT', 'ASC'): 0.00569 0.152 9.85 0 0.00297 0.0911 11 0
('B_DRVRCNT', 'B_DRIVER'): -0.00711 -0.484 -9.61 0 -0.00553 -0.387 -10.1 0
('B_GASPRICE', 'ASC'): -0.000317 -0.914 11.1 0 -0.000241 -0.877 12.7 0
('B_GASPRICE', 'B_DRIVER'): 1.36e-05 0.1 -10 0 1.3e-05 0.108 -10.2 0
('B_GASPRICE', 'B_DRVRCNT'): -9.34e-06 -0.111 5.67 1.39e-08 -2.21e-06 -0.0292 5.72 1.07e-08
('B_HHFAMINC_IMP', 'ASC'): 0.000383 0.0418 11.4 0 0.000689 0.0821 13 0
('B_HHFAMINC_IMP', 'B_DRIVER'): -0.000139 -0.0387 -9.25 0 -1.65e-07 -4.49e-05 -9.49 0
('B_HHFAMINC_IMP', 'B_DRVRCNT'): -0.000257 -0.116 6.3 2.96e-10 -0.000138 -0.0601 6.4 1.53e-10
('B_HHFAMINC_IMP', 'B_GASPRICE'): 1.04e-06 0.0506 4 6.28e-05 1.11e-06 0.0573 3.83 0.000127
('B_HHVEHCNT', 'ASC'): -0.00378 -0.0909 14.4 0 -0.00098 -0.0249 16.4 0
('B_HHVEHCNT', 'B_DRIVER'): 0.00115 0.0707 0.395 0.693 -5.12e-05 -0.00297 0.377 0.706
('B_HHVEHCNT', 'B_DRVRCNT'): -0.006 -0.596 12 0 -0.00583 -0.54 11.8 0
('B_HHVEHCNT', 'B_GASPRICE'): 1.09e-05 0.117 15.3 0 8.94e-06 0.0982 14.2 0
('B_HHVEHCNT', 'B_HHFAMINC_IMP'): -0.000545 -0.221 13.5 0 -0.000628 -0.227 12.5 0
('B_RAIL', 'ASC'): -0.00393 -0.0768 13.6 0 0.0005 0.011 15.6 0
('B_RAIL', 'B_DRIVER'): -0.000491 -0.0245 -0.86 0.39 -0.00118 -0.0591 -0.851 0.395
('B_RAIL', 'B_DRVRCNT'): -0.000663 -0.0535 11.6 0 1.37e-05 0.0011 11.8 0
('B_RAIL', 'B_GASPRICE'): -4.92e-06 -0.0429 10.5 0 -1.41e-05 -0.134 10.4 0
('B_RAIL', 'B_HHFAMINC_IMP'): 0.00118 0.391 10.4 0 0.00119 0.372 10.2 0
('B_RAIL', 'B_HHVEHCNT'): -0.00106 -0.0772 -1.42 0.155 -0.000982 -0.0652 -1.37 0.17
('B_TDWKND', 'ASC'): -0.00327 -0.0483 12.8 0 -0.00798 -0.127 13.7 0
('B_TDWKND', 'B_DRIVER'): 0.000538 0.0202 -1.39 0.163 0.000482 0.0175 -1.36 0.174
('B_TDWKND', 'B_DRVRCNT'): -0.00107 -0.065 8.74 0 -0.00234 -0.136 8.15 4.44e-16
('B_TDWKND', 'B_GASPRICE'): 1.34e-05 0.0882 7.12 1.07e-12 2.3e-05 0.158 6.72 1.85e-11
('B_TDWKND', 'B_HHFAMINC_IMP'): -0.000197 -0.0492 6.48 9.44e-11 0.000148 0.0333 6.18 6.6e-10
('B_TDWKND', 'B_HHVEHCNT'): 0.00239 0.131 -2.06 0.0397 0.00324 0.156 -1.96 0.0505
('B_TDWKND', 'B_RAIL'): -0.000271 -0.0121 -0.662 0.508 0.00269 0.111 -0.674 0.501
('B_WALK', 'ASC'): -0.00109 -0.217 11.5 0 -0.000934 -0.227 13.1 0
('B_WALK', 'B_DRIVER'): -0.000226 -0.114 -8.7 0 -0.000465 -0.258 -8.8 0
('B_WALK', 'B_DRVRCNT'): 9.98e-05 0.0821 7.64 2.15e-14 0.000199 0.177 7.79 6.66e-15
('B_WALK', 'B_GASPRICE'): 4.68e-07 0.0416 14.5 0 2.62e-07 0.0275 15.5 0
('B_WALK', 'B_HHFAMINC_IMP'): 3.49e-05 0.117 3.66 0.000255 4.84e-05 0.167 3.65 0.00026
('B_WALK', 'B_HHVEHCNT'): -0.000224 -0.166 -13.2 0 -0.000542 -0.399 -12 0
('B_WALK', 'B_RAIL'): 6.5e-05 0.0392 -9.11 0 7.5e-05 0.0476 -8.97 0
('B_WALK', 'B_TDWKND'): 1.64e-05 0.00744 -6.03 1.65e-09 -9.56e-05 -0.0439 -5.67 1.43e-08
('B_WORKER', 'ASC'): 0.00771 0.146 9.3 0 0.00426 0.0878 10.2 0
('B_WORKER', 'B_DRIVER'): -0.00432 -0.208 -10 0 -0.00661 -0.311 -9.5 0
('B_WORKER', 'B_DRVRCNT'): 0.00131 0.102 -1.02 0.31 0.0012 0.0901 -0.979 0.328
('B_WORKER', 'B_GASPRICE'): -3.29e-06 -0.0278 -5.22 1.82e-07 2.07e-06 0.0184 -4.97 6.6e-07
('B_WORKER', 'B_HHFAMINC_IMP'): -0.00091 -0.291 -5.56 2.73e-08 -0.000911 -0.266 -5.32 1.04e-07
('B_WORKER', 'B_HHVEHCNT'): 0.000224 0.0158 -13.7 0 -0.000186 -0.0116 -12.7 0
('B_WORKER', 'B_RAIL'): -0.00109 -0.0623 -10.8 0 -0.00108 -0.0582 -10.4 0
('B_WORKER', 'B_TDWKND'): -0.00237 -0.103 -8.41 0 -0.00353 -0.137 -7.85 4.22e-15
('B_WORKER', 'B_WALK'): -5.43e-05 -0.0317 -6.55 5.79e-11 5.49e-05 0.0327 -6.28 3.32e-10
Value Rob. Std err Rob. t-test Rob. p-value
ASC -4.382185 0.344929 -12.704599 0.000000e+00
B_DRIVER 1.550245 0.151089 10.260452 0.000000e+00
B_DRVRCNT -0.536432 0.094601 -5.670494 1.423864e-08
B_GASPRICE 0.004829 0.000799 6.047361 1.472378e-09
B_HHFAMINC_IMP 0.097931 0.024330 4.025211 5.692434e-05
B_HHVEHCNT 1.621649 0.114031 14.221073 0.000000e+00
B_RAIL 1.374609 0.132077 10.407596 0.000000e+00
B_TDWKND 1.230995 0.182671 6.738877 1.596145e-11
B_WALK 0.190136 0.011929 15.938913 0.000000e+00
B_WORKER -0.695368 0.140816 -4.938146 7.886893e-07
Accuracy: 0.9861576570013114
Recall: 0.16091954022988506
Precision: 0.3888888888888889
F1 Score: 0.22764227642276422
ROC AUC Score: 0.5788363934915659
Accuracy: 0.9218241042345277
Recall: 0.181360201511335
Precision: 0.8674698795180723
F1 Score: 0.3
ROC AUC Score: 0.5892702058569235
Accuracy: 0.8301790281329924
Recall: 0.17884130982367757
Precision: 0.922077922077922
F1 Score: 0.29957805907172996
ROC AUC Score: 0.5874951093405936
Accuracy: 0.7520786092214664
Recall: 0.17884130982367757
Precision: 0.9726027397260274
F1 Score: 0.3021276595744681
ROC AUC Score: 0.5883407413049273
Accuracy: 0.6717019133937563
Recall: 0.17884130982367757
Precision: 1.0
F1 Score: 0.3034188034188034
ROC AUC Score: 0.5894206549118388
Accuracy: 0.5826972010178118
Recall: 0.17884130982367757
Precision: 0.9726027397260274
F1 Score: 0.3021276595744681
ROC AUC Score: 0.5868499608244352
#基础模型 probit
#（1）选指标
#（如果要做指标变换，请对训练集和测试集都进行变换！！，并将其新的列名加入positive_features和negative_features）
positive_features = ['GASPRICE', 'RAIL']
negative_features = ['WALK','HHFAMINC_IMP','DRVRCNT','HHVEHCNT','WORKER','TDWKND','DRIVER']
#（2）读训练集
df0 = pd.read_csv('data_train.csv')
#（3）建模（注意类型）
results=build(df0, positive_features, negative_features, 'probit','probit-0')
#（4）获取估计参数
parameters=results.getEstimatedParameters()
beta_values=parameters['Value']
print(parameters)
parameters.to_csv('probit-0-parameters.csv')# 保存参数
#（5）评测
df1=pd.read_csv('data_test.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test1.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test2.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test3.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test4.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test5.csv')# 读测试集
eval(beta_values, positive_features, negative_features, df1)
Results for model probit-0
Output file (HTML): probit-0.html
Nbr of parameters: 10
Sample size: 27448
Excluded data: 0
Init log likelihood: -19025.5
Final log likelihood: -1163.723
Likelihood ratio test (init): 35723.56
Rho square (init): 0.939
Rho bar square (init): 0.938
Akaike Information Criterion: 2347.446
Bayesian Information Criterion: 2429.647
Final gradient norm: 0.04777368
ASC : -2.13[0.186 -11.5 0][0.161 -13.2 0]
B_DRIVER : 0.788[0.0734 10.7 0][0.0704 11.2 0]
B_DRVRCNT : -0.244[0.044 -5.55 2.85e-08][0.0468 -5.22 1.75e-07]
B_GASPRICE : 0.00197[0.000415 4.76 1.93e-06][0.000364 5.42 5.9e-08]
B_HHFAMINC_IMP : 0.0354[0.0112 3.15 0.00162][0.0113 3.13 0.00177]
B_HHVEHCNT : 0.724[0.0481 15 0][0.0545 13.3 0]
B_RAIL : 0.6[0.0603 9.95 0][0.0604 9.93 0]
B_TDWKND : 0.543[0.0786 6.91 4.8e-12][0.0823 6.6 4.06e-11]
B_WALK : 0.097[0.00588 16.5 0][0.00513 18.9 0]
B_WORKER : -0.294[0.0627 -4.7 2.65e-06][0.0658 -4.47 7.69e-06]
('B_DRIVER', 'ASC'): 7.65e-05 0.00562 14.6 0 0.000375 0.033 16.8 0
('B_DRVRCNT', 'ASC'): 0.00138 0.169 10.3 0 0.00076 0.101 11.5 0
('B_DRVRCNT', 'B_DRIVER'): -0.00132 -0.409 -10.3 0 -0.00113 -0.343 -10.6 0
('B_GASPRICE', 'ASC'): -6.99e-05 -0.908 11.5 0 -5.12e-05 -0.872 13.2 0
('B_GASPRICE', 'B_DRIVER'): 2.56e-06 0.084 -10.7 0 2.33e-06 0.0908 -11.2 0
('B_GASPRICE', 'B_DRVRCNT'): -2.17e-06 -0.119 5.59 2.29e-08 -5.56e-07 -0.0326 5.27 1.4e-07
('B_HHFAMINC_IMP', 'ASC'): 0.000109 0.0522 11.7 0 0.0002 0.11 13.5 0
('B_HHFAMINC_IMP', 'B_DRIVER'): -5.97e-05 -0.0724 -10 0 -4.16e-06 -0.00522 -10.5 0
('B_HHFAMINC_IMP', 'B_DRVRCNT'): -5.87e-05 -0.119 5.99 2.11e-09 -4.16e-05 -0.0785 5.71 1.12e-08
('B_HHFAMINC_IMP', 'B_GASPRICE'): 3.3e-07 0.0708 2.98 0.00286 2.93e-07 0.071 2.96 0.00311
('B_HHVEHCNT', 'ASC'): -0.000941 -0.105 14.5 0 -0.000364 -0.0414 16.6 0
('B_HHVEHCNT', 'B_DRIVER'): 0.00014 0.0395 -0.748 0.455 -0.000133 -0.0347 -0.712 0.477
('B_HHVEHCNT', 'B_DRVRCNT'): -0.0013 -0.613 11.7 0 -0.00143 -0.561 10.8 0
('B_HHVEHCNT', 'B_GASPRICE'): 2.2e-06 0.11 15 0 1.75e-06 0.0882 13.3 0
('B_HHVEHCNT', 'B_HHFAMINC_IMP'): -0.000125 -0.23 13.3 0 -9.73e-05 -0.158 12 0
('B_RAIL', 'ASC'): -0.000717 -0.0641 13.7 0 -0.000161 -0.0165 15.8 0
('B_RAIL', 'B_DRIVER'): -0.000163 -0.0368 -1.95 0.0518 -0.000299 -0.0704 -1.96 0.05
('B_RAIL', 'B_DRVRCNT'): -0.000146 -0.0549 11 0 5.44e-05 0.0193 11.2 0
('B_RAIL', 'B_GASPRICE'): -1.02e-06 -0.0406 9.91 0 -1.57e-06 -0.0712 9.89 0
('B_RAIL', 'B_HHFAMINC_IMP'): 0.000248 0.366 9.88 0 0.00025 0.365 9.86 0
('B_RAIL', 'B_HHVEHCNT'): -0.000216 -0.0745 -1.55 0.122 9.58e-05 0.0291 -1.54 0.123
('B_TDWKND', 'ASC'): -0.00024 -0.0164 13.2 0 -0.00129 -0.0973 14.2 0
('B_TDWKND', 'B_DRIVER'): 0.000167 0.029 -2.31 0.0207 5.47e-05 0.00945 -2.27 0.023
('B_TDWKND', 'B_DRVRCNT'): -0.000206 -0.0595 8.53 0 -0.000118 -0.0308 8.21 2.22e-16
('B_TDWKND', 'B_GASPRICE'): 2.36e-06 0.0724 6.89 5.63e-12 4.77e-06 0.159 6.58 4.63e-11
('B_TDWKND', 'B_HHFAMINC_IMP'): -3.97e-05 -0.045 6.36 2.08e-10 3.93e-05 0.0422 6.15 7.82e-10
('B_TDWKND', 'B_HHVEHCNT'): 0.000533 0.141 -2.1 0.0361 0.000552 0.123 -1.94 0.0519
('B_TDWKND', 'B_RAIL'): -5.5e-05 -0.0116 -0.572 0.567 0.000668 0.134 -0.598 0.55
('B_WALK', 'ASC'): -0.000212 -0.195 11.9 0 -0.000176 -0.213 13.7 0
('B_WALK', 'B_DRIVER'): -4.56e-05 -0.106 -9.31 0 -8.79e-05 -0.244 -9.62 0
('B_WALK', 'B_DRVRCNT'): 9.23e-06 0.0357 7.72 1.15e-14 3.76e-05 0.157 7.38 1.58e-13
('B_WALK', 'B_GASPRICE'): 1.37e-07 0.0561 16.2 0 1.34e-07 0.0718 18.6 0
('B_WALK', 'B_HHFAMINC_IMP'): 7.42e-06 0.112 5.09 3.49e-07 7.99e-06 0.137 5.22 1.75e-07
('B_WALK', 'B_HHVEHCNT'): -1.24e-05 -0.0439 -12.9 0 -7.24e-05 -0.259 -11.2 0
('B_WALK', 'B_RAIL'): 1.52e-05 0.0428 -8.34 0 8.55e-06 0.0276 -8.31 0
('B_WALK', 'B_TDWKND'): 1.06e-05 0.0229 -5.67 1.42e-08 -2.22e-06 -0.00526 -5.41 6.29e-08
('B_WORKER', 'ASC'): 0.00183 0.157 9.85 0 0.00116 0.109 11 0
('B_WORKER', 'B_DRIVER'): -0.000938 -0.204 -10.2 0 -0.00132 -0.285 -9.91 0
('B_WORKER', 'B_DRVRCNT'): 0.000289 0.105 -0.686 0.493 0.000135 0.0437 -0.631 0.528
('B_WORKER', 'B_GASPRICE'): -1.09e-06 -0.0421 -4.73 2.28e-06 -2.96e-07 -0.0124 -4.5 6.7e-06
('B_WORKER', 'B_HHFAMINC_IMP'): -0.000165 -0.234 -4.98 6.34e-07 -0.000171 -0.229 -4.76 1.93e-06
('B_WORKER', 'B_HHVEHCNT'): -4.06e-05 -0.0134 -12.8 0 -0.000258 -0.072 -11.5 0
('B_WORKER', 'B_RAIL'): -0.000167 -0.0441 -10.1 0 -0.000483 -0.121 -9.46 0
('B_WORKER', 'B_TDWKND'): -0.000429 -0.0872 -8 1.33e-15 -0.000804 -0.149 -7.43 1.09e-13
('B_WORKER', 'B_WALK'): -8.12e-06 -0.0221 -6.2 5.48e-10 1.47e-05 0.0434 -5.95 2.68e-09
Value Rob. Std err Rob. t-test Rob. p-value
ASC -2.127947 0.161170 -13.203085 0.000000e+00
B_DRIVER 0.788024 0.070389 11.195310 0.000000e+00
B_DRVRCNT -0.244436 0.046787 -5.224482 1.746436e-07
B_GASPRICE 0.001974 0.000364 5.421785 5.900680e-08
B_HHFAMINC_IMP 0.035436 0.011338 3.125546 1.774752e-03
B_HHVEHCNT 0.723612 0.054489 13.279995 0.000000e+00
B_RAIL 0.599969 0.060428 9.928665 0.000000e+00
B_TDWKND 0.543018 0.082251 6.601947 4.057932e-11
B_WALK 0.097004 0.005130 18.908240 0.000000e+00
B_WORKER -0.294281 0.065783 -4.473518 7.694306e-06
Accuracy: 0.9870319102433338
Recall: 0.14942528735632185
Precision: 0.4642857142857143
F1 Score: 0.22608695652173913
ROC AUC Score: 0.5736057959804042
Accuracy: 0.9192647743136343
Recall: 0.1486146095717884
Precision: 0.8676470588235294
F1 Score: 0.2537634408602151
ROC AUC Score: 0.5731537544141946
Accuracy: 0.8245524296675192
Recall: 0.14609571788413098
Precision: 0.9354838709677419
F1 Score: 0.25272331154684097
ROC AUC Score: 0.5717641618945686
Accuracy: 0.7430083144368859
Recall: 0.14609571788413098
Precision: 0.9830508474576272
F1 Score: 0.2543859649122807
ROC AUC Score: 0.5725079021386098
Accuracy: 0.6586102719033232
Recall: 0.14609571788413098
Precision: 1.0
F1 Score: 0.2549450549450549
ROC AUC Score: 0.5730478589420654
Accuracy: 0.5661577608142494
Recall: 0.14609571788413098
Precision: 0.9666666666666667
F1 Score: 0.2538293216630197
ROC AUC Score: 0.5704771648546618
2. 进阶模型
########logit-advance
#（在基础Logit上新增平均收入）
#从第一部分的Logit模型建立情况来看，变量NUMONTRP,HHFAMINC_IMP,URBRUR的关联性并不显著
#（1）选指标
#（如果要做指标变换，请对训练集和测试集都进行变换！！并将其新的列名加入positive_features和negative_features）
positive_features = ['GASPRICE', 'RAIL']
negative_features = ['WALK','AVERAGE_HHFAMINC_IMP','DRVRCNT','HHVEHCNT','WORKER','TDWKND','DRIVER']
#（2）读训练集
df0 = pd.read_csv('data_train.csv')
#对于HHFAMINC_IMP，我们选择以下处理：
df0['AVERAGE_HHFAMINC_IMP']=df0['HHFAMINC_IMP']/df0['HHSIZE']
#（3）建模结果
results=build(df0, positive_features, negative_features, 'logit','logit-advance')
#（4）获取估计参数
parameters=results.getEstimatedParameters()
print(parameters)
beta_values=parameters['Value']
parameters.to_csv('logit-advance-parameters.csv')# 保存参数
#（5）评测
df1=pd.read_csv('data_test.csv')
df1['AVERAGE_HHFAMINC_IMP']=df1['HHFAMINC_IMP']/df1['HHSIZE']
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test1.csv')
df1['AVERAGE_HHFAMINC_IMP']=df1['HHFAMINC_IMP']/df1['HHSIZE']
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test2.csv')
df1['AVERAGE_HHFAMINC_IMP']=df1['HHFAMINC_IMP']/df1['HHSIZE']
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test3.csv')
df1['AVERAGE_HHFAMINC_IMP']=df1['HHFAMINC_IMP']/df1['HHSIZE']
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test4.csv')
df1['AVERAGE_HHFAMINC_IMP']=df1['HHFAMINC_IMP']/df1['HHSIZE']
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test5.csv')
df1['AVERAGE_HHFAMINC_IMP']=df1['HHFAMINC_IMP']/df1['HHSIZE']
eval(beta_values, positive_features, negative_features, df1)
Results for model logit-advance
Output file (HTML): logit-advance.html
Nbr of parameters: 10
Sample size: 27448
Excluded data: 0
Init log likelihood: -19025.5
Final log likelihood: -1168.625
Likelihood ratio test (init): 35713.76
Rho square (init): 0.939
Rho bar square (init): 0.938
Akaike Information Criterion: 2357.249
Bayesian Information Criterion: 2439.45
Final gradient norm: 0.001179723
ASC : -4.33[0.394 -11 0][0.345 -12.5 0]
B_AVERAGE_HHFAMINC_IMP: 0.137[0.0302 4.54 5.55e-06][0.0329 4.18 2.96e-05]
B_DRIVER : 1.36[0.163 8.34 0][0.16 8.51 0]
B_DRVRCNT : -0.352[0.101 -3.49 0.000476][0.105 -3.36 0.000792]
B_GASPRICE : 0.0048[0.000884 5.43 5.68e-08][0.000805 5.96 2.46e-09]
B_HHVEHCNT : 1.66[0.104 15.9 0][0.112 14.8 0]
B_RAIL : 1.33[0.125 10.7 0][0.126 10.6 0]
B_TDWKND : 1.15[0.173 6.64 3.06e-11][0.181 6.33 2.4e-10]
B_WALK : 0.188[0.0128 14.8 0][0.0118 15.9 0]
B_WORKER : -0.642[0.131 -4.91 8.94e-07][0.139 -4.62 3.93e-06]
('B_AVERAGE_HHFAMINC_IMP', 'ASC'): 0.000732 0.0614 11.4 0 0.000901 0.0793 13 0
('B_DRIVER', 'ASC'): -0.00282 -0.044 13.1 0 -0.000508 -0.00923 14.9 0
('B_DRIVER', 'B_AVERAGE_HHFAMINC_IMP'): -0.00129 -0.262 7.05 1.84e-12 -0.00113 -0.214 7.19 6.41e-13
('B_DRVRCNT', 'ASC'): 0.00624 0.157 10.2 0 0.00357 0.0984 11.3 0
('B_DRVRCNT', 'B_AVERAGE_HHFAMINC_IMP'): 0.000776 0.254 -5.02 5.27e-07 0.00103 0.297 -4.88 1.04e-06
('B_DRVRCNT', 'B_DRIVER'): -0.0089 -0.542 -7.33 2.36e-13 -0.0078 -0.466 -7.49 6.75e-14
('B_GASPRICE', 'ASC'): -0.000318 -0.913 11 0 -0.000243 -0.875 12.5 0
('B_GASPRICE', 'B_AVERAGE_HHFAMINC_IMP'): 1.21e-06 0.0452 -4.39 1.14e-05 2.3e-06 0.087 -4.04 5.39e-05
('B_GASPRICE', 'B_DRIVER'): 1.15e-05 0.0798 -8.31 0 9.08e-06 0.0707 -8.48 0
('B_GASPRICE', 'B_DRVRCNT'): -7.07e-06 -0.0793 3.54 0.000402 1.69e-06 0.0199 3.4 0.00067
('B_HHVEHCNT', 'ASC'): -0.00389 -0.0948 14.4 0 -0.000791 -0.0204 16.4 0
('B_HHVEHCNT', 'B_AVERAGE_HHFAMINC_IMP'): -0.000355 -0.113 13.6 0 -0.000376 -0.102 12.7 0
('B_HHVEHCNT', 'B_DRIVER'): 0.0017 0.1 1.64 0.101 0.000499 0.0279 1.57 0.117
('B_HHVEHCNT', 'B_DRVRCNT'): -0.0068 -0.648 10.8 0 -0.00675 -0.574 10.4 0
('B_HHVEHCNT', 'B_GASPRICE'): 1.17e-05 0.127 15.9 0 9.14e-06 0.101 14.8 0
('B_RAIL', 'ASC'): -0.00387 -0.0787 13.4 0 0.00012 0.00276 15.4 0
('B_RAIL', 'B_AVERAGE_HHFAMINC_IMP'): 0.00109 0.288 9.96 0 0.000941 0.228 9.76 0
('B_RAIL', 'B_DRIVER'): -0.00195 -0.096 -0.124 0.902 -0.00256 -0.128 -0.123 0.902
('B_RAIL', 'B_DRVRCNT'): 0.00106 0.0843 10.9 0 0.00168 0.127 11 0
('B_RAIL', 'B_GASPRICE'): -5.55e-06 -0.0502 10.6 0 -1.41e-05 -0.139 10.6 0
('B_RAIL', 'B_HHVEHCNT'): -0.000375 -0.0288 -1.99 0.0464 -0.000141 -0.01 -1.94 0.052
('B_TDWKND', 'ASC'): -0.00341 -0.0501 12.5 0 -0.00814 -0.13 13.3 0
('B_TDWKND', 'B_AVERAGE_HHFAMINC_IMP'): -0.000735 -0.141 5.63 1.82e-08 -0.000877 -0.147 5.35 8.84e-08
('B_TDWKND', 'B_DRIVER'): 0.00151 0.0537 -0.909 0.363 0.00164 0.0566 -0.895 0.371
('B_TDWKND', 'B_DRVRCNT'): -0.00182 -0.105 7.18 7e-13 -0.00313 -0.165 6.7 2.09e-11
('B_TDWKND', 'B_GASPRICE'): 1.25e-05 0.0818 6.62 3.62e-11 2.02e-05 0.138 6.31 2.78e-10
('B_TDWKND', 'B_HHVEHCNT'): 0.00241 0.134 -2.7 0.00686 0.0035 0.172 -2.61 0.00897
('B_TDWKND', 'B_RAIL'): -0.000878 -0.0407 -0.844 0.398 0.00113 0.0498 -0.852 0.394
('B_WALK', 'ASC'): -0.00108 -0.215 11.4 0 -0.000888 -0.218 13 0
('B_WALK', 'B_AVERAGE_HHFAMINC_IMP'): 2.91e-05 0.0754 1.6 0.111 2.69e-05 0.0693 1.49 0.136
('B_WALK', 'B_DRIVER'): -0.000253 -0.122 -7.09 1.3e-12 -0.00049 -0.26 -7.17 7.38e-13
('B_WALK', 'B_DRVRCNT'): 0.000142 0.11 5.39 6.9e-08 0.000252 0.203 5.24 1.64e-07
('B_WALK', 'B_GASPRICE'): 4.03e-07 0.0358 14.4 0 4.49e-08 0.00473 15.5 0
('B_WALK', 'B_HHVEHCNT'): -0.000199 -0.15 -13.8 0 -0.000501 -0.379 -12.6 0
('B_WALK', 'B_RAIL'): 2.96e-05 0.0186 -9.12 0 1.48e-05 0.01 -9.07 0
('B_WALK', 'B_TDWKND'): -8.8e-06 -0.004 -5.54 3.08e-08 -0.000133 -0.0622 -5.26 1.43e-07
('B_WORKER', 'ASC'): 0.00774 0.15 9.31 0 0.00429 0.0892 10.2 0
('B_WORKER', 'B_AVERAGE_HHFAMINC_IMP'): -0.000716 -0.181 -5.59 2.23e-08 -0.000984 -0.215 -5.21 1.92e-07
('B_WORKER', 'B_DRIVER'): -0.00372 -0.175 -8.85 0 -0.00578 -0.26 -8.42 0
('B_WORKER', 'B_DRVRCNT'): 0.000357 0.0271 -1.78 0.0751 0.000292 0.02 -1.68 0.0931
('B_WORKER', 'B_GASPRICE'): -2.55e-06 -0.0221 -4.95 7.44e-07 2.45e-06 0.0219 -4.65 3.32e-06
('B_WORKER', 'B_HHVEHCNT'): -0.000454 -0.0334 -13.6 0 -0.000855 -0.0549 -12.6 0
('B_WORKER', 'B_RAIL'): 0.000124 0.00759 -11 0 1.52e-05 0.000871 -10.5 0
('B_WORKER', 'B_TDWKND'): -0.00197 -0.0871 -7.94 2e-15 -0.00223 -0.0884 -7.52 5.48e-14
('B_WORKER', 'B_WALK'): -1.95e-05 -0.0117 -6.32 2.66e-10 0.000101 0.0613 -5.98 2.26e-09
Value Rob. Std err Rob. t-test Rob. p-value
ASC -4.329425 0.345295 -12.538340 0.000000e+00
B_AVERAGE_HHFAMINC_IMP 0.137402 0.032898 4.176573 2.959340e-05
B_DRIVER 1.357883 0.159591 8.508525 0.000000e+00
B_DRVRCNT -0.352331 0.105004 -3.355401 7.924989e-04
B_GASPRICE 0.004801 0.000805 5.964163 2.458917e-09
B_HHVEHCNT 1.659870 0.112025 14.817007 0.000000e+00
B_RAIL 1.331328 0.125557 10.603408 0.000000e+00
B_TDWKND 1.147835 0.181234 6.333447 2.397440e-10
B_WALK 0.188352 0.011813 15.944949 0.000000e+00
B_WORKER -0.642287 0.139173 -4.615024 3.930493e-06
Accuracy: 0.986740492495993
Recall: 0.16091954022988506
Precision: 0.4375
F1 Score: 0.23529411764705882
ROC AUC Score: 0.5791315528776344
Accuracy: 0.9222894369474174
Recall: 0.181360201511335
Precision: 0.8888888888888888
F1 Score: 0.301255230125523
ROC AUC Score: 0.589526550383968
Accuracy: 0.8317135549872122
Recall: 0.17884130982367757
Precision: 0.9594594594594594
F1 Score: 0.30148619957537154
ROC AUC Score: 0.5884578821262162
Accuracy: 0.7528344671201814
Recall: 0.17884130982367757
Precision: 0.9861111111111112
F1 Score: 0.302771855010661
ROC AUC Score: 0.5888806981083831
Accuracy: 0.6717019133937563
Recall: 0.17884130982367757
Precision: 1.0
F1 Score: 0.3034188034188034
ROC AUC Score: 0.5894206549118388
Accuracy: 0.5826972010178118
Recall: 0.17884130982367757
Precision: 0.9726027397260274
F1 Score: 0.3021276595744681
ROC AUC Score: 0.5868499608244352
########probit-advance
#（1）选指标
#（如果要做指标变换，请对训练集和测试集都进行变换！！并将其新的列名加入positive_features和negative_features）
positive_features = ['GASPRICE', 'RAIL']
negative_features = ['WALK','AVERAGE_HHFAMINC_IMP','DRVRCNT','HHVEHCNT','WORKER','TDWKND','DRIVER']
#（2）读训练集
df0 = pd.read_csv('data_train.csv')
#对于HHFAMINC_IMP，我们选择以下处理：
df0['AVERAGE_HHFAMINC_IMP']=df0['HHFAMINC_IMP']/df0['HHSIZE']
#（3）建模结果
results=build(df0, positive_features, negative_features, 'probit','probit-1')
#（4）获取估计参数
parameters=results.getEstimatedParameters()
print(parameters)
beta_values=parameters['Value']
parameters.to_csv('probit-1-parameters.csv')# 保存参数
#（5）评测
df1=pd.read_csv('data_test.csv')
df1['AVERAGE_HHFAMINC_IMP']=df1['HHFAMINC_IMP']/df1['HHSIZE']
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test1.csv')
df1['AVERAGE_HHFAMINC_IMP']=df1['HHFAMINC_IMP']/df1['HHSIZE']
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test2.csv')
df1['AVERAGE_HHFAMINC_IMP']=df1['HHFAMINC_IMP']/df1['HHSIZE']
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test3.csv')
df1['AVERAGE_HHFAMINC_IMP']=df1['HHFAMINC_IMP']/df1['HHSIZE']
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test4.csv')
df1['AVERAGE_HHFAMINC_IMP']=df1['HHFAMINC_IMP']/df1['HHSIZE']
eval(beta_values, positive_features, negative_features, df1)
df1=pd.read_csv('data_test5.csv')
df1['AVERAGE_HHFAMINC_IMP']=df1['HHFAMINC_IMP']/df1['HHSIZE']
eval(beta_values, positive_features, negative_features, df1)
Results for model probit-1
Output file (HTML): probit-1.html
Nbr of parameters: 10
Sample size: 27448
Excluded data: 0
Init log likelihood: -19025.5
Final log likelihood: -1161.982
Likelihood ratio test (init): 35727.04
Rho square (init): 0.939
Rho bar square (init): 0.938
Akaike Information Criterion: 2343.964
Bayesian Information Criterion: 2426.165
Final gradient norm: 0.04058286
ASC : -2.08[0.186 -11.2 0][0.162 -12.8 0]
B_AVERAGE_HHFAMINC_IMP: 0.0526[0.0145 3.62 0.000299][0.0149 3.52 0.00043]
B_DRIVER : 0.71[0.0779 9.11 0][0.0752 9.44 0]
B_DRVRCNT : -0.174[0.0466 -3.73 0.000191][0.0514 -3.38 0.00072]
B_GASPRICE : 0.00192[0.000414 4.64 3.5e-06][0.000364 5.27 1.33e-07]
B_HHVEHCNT : 0.737[0.0472 15.6 0][0.054 13.6 0]
B_RAIL : 0.589[0.0583 10.1 0][0.058 10.2 0]
B_TDWKND : 0.521[0.0788 6.61 3.91e-11][0.0815 6.38 1.72e-10]
B_WALK : 0.0962[0.00586 16.4 0][0.0051 18.9 0]
B_WORKER : -0.276[0.0614 -4.49 7.07e-06][0.0653 -4.22 2.41e-05]
('B_AVERAGE_HHFAMINC_IMP', 'ASC'): 0.000288 0.106 11.5 0 0.000332 0.137 13.3 0
('B_DRIVER', 'ASC'): -0.000361 -0.0248 13.7 0 -3.19e-05 -0.00262 15.6 0
('B_DRIVER', 'B_AVERAGE_HHFAMINC_IMP'): -0.000358 -0.316 7.86 3.77e-15 -0.0003 -0.268 8.17 2.22e-16
('B_DRVRCNT', 'ASC'): 0.00169 0.195 10.4 0 0.00114 0.136 11.7 0
('B_DRVRCNT', 'B_AVERAGE_HHFAMINC_IMP'): 0.0002 0.296 -5.09 3.65e-07 0.000246 0.321 -4.65 3.35e-06
('B_DRVRCNT', 'B_DRIVER'): -0.00177 -0.487 -8.15 4.44e-16 -0.00168 -0.435 -8.19 2.22e-16
('B_GASPRICE', 'ASC'): -7e-05 -0.906 11.2 0 -5.14e-05 -0.87 12.8 0
('B_GASPRICE', 'B_AVERAGE_HHFAMINC_IMP'): 1.82e-07 0.0302 -3.49 0.000492 3.16e-07 0.058 -3.4 0.000684
('B_GASPRICE', 'B_DRIVER'): 2.34e-06 0.0723 -9.09 0 1.6e-06 0.0584 -9.42 0
('B_GASPRICE', 'B_DRVRCNT'): -1.75e-06 -0.0909 3.77 0.000164 1.8e-07 0.00962 3.42 0.000627
('B_HHVEHCNT', 'ASC'): -0.000954 -0.108 14.3 0 -0.000237 -0.027 16.4 0
('B_HHVEHCNT', 'B_AVERAGE_HHFAMINC_IMP'): -8.05e-05 -0.117 13.4 0 -4.37e-05 -0.0542 12 0
('B_HHVEHCNT', 'B_DRIVER'): 0.000236 0.064 0.302 0.763 -7.15e-05 -0.0176 0.286 0.775
('B_HHVEHCNT', 'B_DRVRCNT'): -0.00146 -0.663 10.6 0 -0.00156 -0.564 9.77 0
('B_HHVEHCNT', 'B_GASPRICE'): 2.44e-06 0.125 15.6 0 1.7e-06 0.0862 13.6 0
('B_RAIL', 'ASC'): -0.00062 -0.057 13.5 0 -0.000265 -0.0282 15.4 0
('B_RAIL', 'B_AVERAGE_HHFAMINC_IMP'): 0.000231 0.272 9.56 0 0.00021 0.242 9.53 0
('B_RAIL', 'B_DRIVER'): -0.000456 -0.1 -1.19 0.234 -0.000639 -0.147 -1.19 0.232
('B_RAIL', 'B_DRVRCNT'): 0.000208 0.0766 10.6 0 0.000442 0.148 10.7 0
('B_RAIL', 'B_GASPRICE'): -1.44e-06 -0.0597 10.1 0 -1.68e-06 -0.0795 10.1 0
('B_RAIL', 'B_HHVEHCNT'): -6.3e-05 -0.0228 -1.95 0.0513 0.000238 0.0761 -1.94 0.0522
('B_TDWKND', 'ASC'): -0.000359 -0.0244 12.8 0 -0.00138 -0.104 13.8 0
('B_TDWKND', 'B_AVERAGE_HHFAMINC_IMP'): -0.00013 -0.113 5.73 1.02e-08 -8.97e-05 -0.0737 5.57 2.49e-08
('B_TDWKND', 'B_DRIVER'): 0.000363 0.0591 -1.76 0.0776 0.000198 0.0323 -1.74 0.0821
('B_TDWKND', 'B_DRVRCNT'): -0.000349 -0.0951 7.29 3.12e-13 -0.000171 -0.0407 7.08 1.49e-12
('B_TDWKND', 'B_GASPRICE'): 2.34e-06 0.0716 6.59 4.54e-11 4.43e-06 0.149 6.36 1.95e-10
('B_TDWKND', 'B_HHVEHCNT'): 0.000531 0.143 -2.52 0.0117 0.000555 0.126 -2.35 0.0186
('B_TDWKND', 'B_RAIL'): -0.000138 -0.0301 -0.689 0.491 0.00047 0.0994 -0.719 0.472
('B_WALK', 'ASC'): -0.000213 -0.195 11.6 0 -0.000178 -0.215 13.3 0
('B_WALK', 'B_AVERAGE_HHFAMINC_IMP'): 5.61e-06 0.0658 2.85 0.00433 5.14e-06 0.0674 2.83 0.00468
('B_WALK', 'B_DRIVER'): -5.01e-05 -0.11 -7.79 6.66e-15 -9.42e-05 -0.246 -8.01 1.11e-15
('B_WALK', 'B_DRVRCNT'): 1.75e-05 0.0643 5.8 6.68e-09 4.78e-05 0.182 5.33 1e-07
('B_WALK', 'B_GASPRICE'): 1.24e-07 0.051 16.1 0 1.16e-07 0.0622 18.5 0
('B_WALK', 'B_HHVEHCNT'): -7.75e-06 -0.028 -13.4 0 -6.88e-05 -0.25 -11.5 0
('B_WALK', 'B_RAIL'): 7.43e-06 0.0217 -8.42 0 4.34e-08 0.000147 -8.46 0
('B_WALK', 'B_TDWKND'): 7.11e-06 0.0154 -5.38 7.56e-08 -8.26e-06 -0.0199 -5.19 2.13e-07
('B_WORKER', 'ASC'): 0.00182 0.159 9.68 0 0.00116 0.109 10.8 0
('B_WORKER', 'B_AVERAGE_HHFAMINC_IMP'): -0.00011 -0.124 -5.07 4.08e-07 -0.000188 -0.193 -4.71 2.5e-06
('B_WORKER', 'B_DRIVER'): -0.000852 -0.178 -9.18 0 -0.00111 -0.225 -8.95 0
('B_WORKER', 'B_DRVRCNT'): 0.000124 0.0432 -1.35 0.176 -7.75e-05 -0.0231 -1.22 0.224
('B_WORKER', 'B_GASPRICE'): -7.78e-07 -0.0306 -4.52 6.13e-06 -9.43e-08 -0.00396 -4.25 2.12e-05
('B_WORKER', 'B_HHVEHCNT'): -0.000172 -0.0594 -12.7 0 -0.000352 -0.0998 -11.4 0
('B_WORKER', 'B_RAIL'): 4.22e-05 0.0118 -10.3 0 -0.000326 -0.086 -9.5 0
('B_WORKER', 'B_TDWKND'): -0.000395 -0.0816 -7.67 1.67e-14 -0.00066 -0.124 -7.2 6.03e-13
('B_WORKER', 'B_WALK'): -1.25e-06 -0.00348 -6.03 1.65e-09 2.12e-05 0.0635 -5.71 1.15e-08
Value Rob. Std err Rob. t-test Rob. p-value
ASC -2.083226 0.162199 -12.843647 0.000000e+00
B_AVERAGE_HHFAMINC_IMP 0.052551 0.014925 3.521051 4.298397e-04
B_DRIVER 0.710254 0.075203 9.444537 0.000000e+00
B_DRVRCNT -0.173726 0.051369 -3.381936 7.197684e-04
B_GASPRICE 0.001922 0.000364 5.274059 1.334391e-07
B_HHVEHCNT 0.736994 0.054004 13.647136 0.000000e+00
B_RAIL 0.589076 0.058005 10.155702 0.000000e+00
B_TDWKND 0.520549 0.081535 6.384344 1.721332e-10
B_WALK 0.096229 0.005103 18.856777 0.000000e+00
B_WORKER -0.275877 0.065331 -4.222763 2.413251e-05
Accuracy: 0.9871776191170042
Recall: 0.14942528735632185
Precision: 0.48148148148148145
F1 Score: 0.22807017543859648
ROC AUC Score: 0.5736795858269211
Accuracy: 0.9194974406700791
Recall: 0.1486146095717884
Precision: 0.8805970149253731
F1 Score: 0.2543103448275862
ROC AUC Score: 0.5732819266777168
Accuracy: 0.8250639386189258
Recall: 0.14609571788413098
Precision: 0.9508196721311475
F1 Score: 0.25327510917030566
ROC AUC Score: 0.5720850861564429
Accuracy: 0.7430083144368859
Recall: 0.14609571788413098
Precision: 0.9830508474576272
F1 Score: 0.2543859649122807
ROC AUC Score: 0.5725079021386098
Accuracy: 0.6586102719033232
Recall: 0.14609571788413098
Precision: 1.0
F1 Score: 0.2549450549450549
ROC AUC Score: 0.5730478589420654
Accuracy: 0.5661577608142494
Recall: 0.14609571788413098
Precision: 0.9666666666666667
F1 Score: 0.2538293216630197
ROC AUC Score: 0.5704771648546618
STEP3 机器学习模型——XGBoost
1. XGBoost建模
import shap
import xgboost as xgb
from xgboost import plot_importance, XGBClassifier
import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns
from sklearn.model_selection import GridSearchCV, train_test_split, cross_val_score
from sklearn.metrics import accuracy_score, recall_score, precision_score, f1_score, roc_auc_score, confusion_matrix
from sklearn.preprocessing import OneHotEncoder, StandardScaler
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
def calculate_metrics(y_test, y_pred):
accuracy = accuracy_score(y_test, y_pred)
recall = recall_score(y_test, y_pred)
precision = precision_score(y_test, y_pred)
f1 = f1_score(y_test, y_pred)
roc_auc = roc_auc_score(y_test, y_pred)
cm = confusion_matrix(y_test, y_pred)
print(f"Accuracy: {accuracy}")
print(f"Recall: {recall}")
print(f"Precision: {precision}")
print(f"F1 Score: {f1}")
print(f"ROC AUC Score: {roc_auc}")
# 绘制混淆矩阵的热力图
plt.figure(figsize=(10, 7))
sns.heatmap(cm, annot=True, fmt='d', cmap='Blues')
plt.xlabel('Predicted')
plt.ylabel('True')
plt.show()
# 分类变量和连续变量
categorical_features = ['WORKER', 'URBRUR', 'TRIPPURP', 'TDWKND', 'DRIVER', 'RAIL']
numeric_features = ['WALK', 'TRPMILES', 'HHFAMINC_IMP', 'NUMONTRP', 'GASPRICE', 'DRVRCNT', 'HHVEHCNT', 'HHSIZE','R_AGE']
# 导入扩样后数据
df_rep0=pd.read_csv('data_whole.csv')
# (1)如果要每次重新shuffle训练集、测试集：
#X_train, X_test, y_train, y_test = train_test_split(df_rep0[categorical_features + numeric_features], df_rep0['PUBTRANS'].values, test_size=0.2, random_state=42, shuffle=True)
# (2)如果每次用一样的
X_train=pd.read_csv('data_train.csv')
y_train = X_train['PUBTRANS'].values
X_train = X_train.drop('PUBTRANS', axis=1)
# 创建预处理器
preprocessor = ColumnTransformer(
transformers=[
('num', StandardScaler(), numeric_features),
('cat', OneHotEncoder(), categorical_features)])
# 创建分类器
clf = XGBClassifier(
booster='gbtree',
n_estimators=100,
learning_rate=0.1,
reg_alpha=0,
reg_lambda=1,
subsample=1,
colsample_bytree=1,
colsample_bynode=1,
min_child_weight=0,
max_depth=10,
random_state=42
)
# 创建管道
pipe = Pipeline(steps=[('preprocessor', preprocessor),
('classifier', clf)])
(1) 使用网格搜索优化超参数
# 定义参数网格
param_grid = {
'classifier__n_estimators': [50, 100, 150, 200, 250],
'classifier__learning_rate': [0.01, 0.05, 0.1, 0.2, 0.3],
'classifier__max_depth': [4, 6, 8, 10, 12]
}
# 创建网格搜索对象，使用交叉验证
grid_search = GridSearchCV(pipe, param_grid, cv=5, verbose=3)
# 执行网格搜索
grid_search.fit(df_rep0[categorical_features + numeric_features], df_rep0['PUBTRANS'].values)
# 输出最佳参数和对应的评分
print(f"Best parameters: {grid_search.best_params_}")
print(f"Best cross-validation score: {grid_search.best_score_}")
# 使用最佳参数创建新的管道
best_clf = grid_search.best_estimator_
# 训练模型
best_clf.fit(X_train, y_train)
# 预测
X_test=pd.read_csv('data_test.csv')
y_test = X_test['PUBTRANS'].values
X_test = X_test.drop('PUBTRANS', axis=1)
y_pred = best_clf.predict(X_test)
calculate_metrics(y_test, y_pred)
X_test=pd.read_csv('data_test2.csv')
y_test = X_test['PUBTRANS'].values
X_test = X_test.drop('PUBTRANS', axis=1)
y_pred = best_clf.predict(X_test)
calculate_metrics(y_test, y_pred)
(2) 使用交叉验证
# 使用交叉验证评估模型
scores = cross_val_score(pipe, df_rep0[categorical_features + numeric_features], df_rep0['PUBTRANS'].values, cv=5)
print(f"Cross-validation scores: {scores}")
print(f"Average cross-validation score: {scores.mean()}")
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score, recall_score, precision_score, f1_score, roc_auc_score, confusion_matrix
import seaborn as sns
import matplotlib.pyplot as plt
# 训练模型
pipe.fit(X_train, y_train)
# 预测
X_test=pd.read_csv('data_test.csv')
y_test = X_test['PUBTRANS'].values
X_test = X_test.drop('PUBTRANS', axis=1)
y_pred = pipe.predict(X_test)
calculate_metrics(y_test, y_pred)
X_test=pd.read_csv('data_test1.csv')
y_test = X_test['PUBTRANS'].values
X_test = X_test.drop('PUBTRANS', axis=1)
y_pred = pipe.predict(X_test)
calculate_metrics(y_test, y_pred)
X_test=pd.read_csv('data_test2.csv')
y_test = X_test['PUBTRANS'].values
X_test = X_test.drop('PUBTRANS', axis=1)
y_pred = pipe.predict(X_test)
calculate_metrics(y_test, y_pred)
X_test=pd.read_csv('data_test3.csv')
y_test = X_test['PUBTRANS'].values
X_test = X_test.drop('PUBTRANS', axis=1)
y_pred = pipe.predict(X_test)
calculate_metrics(y_test, y_pred)
X_test=pd.read_csv('data_test4.csv')
y_test = X_test['PUBTRANS'].values
X_test = X_test.drop('PUBTRANS', axis=1)
y_pred = pipe.predict(X_test)
calculate_metrics(y_test, y_pred)
X_test=pd.read_csv('data_test5.csv')
y_test = X_test['PUBTRANS'].values
X_test = X_test.drop('PUBTRANS', axis=1)
y_pred = pipe.predict(X_test)
calculate_metrics(y_test, y_pred)
Cross-validation scores: [0.98936325 0.9864471 0.99023608 0.98571845 0.99009035]
Average cross-validation score: 0.9883710474132258
Accuracy: 0.9957744426635582
Recall: 0.8045977011494253
Precision: 0.8536585365853658
F1 Score: 0.8284023668639053
ROC AUC Score: 0.9014133724165072
Accuracy: 0.9946486738017682
Recall: 0.9622166246851386
Precision: 0.9794871794871794
F1 Score: 0.9707750952986023
ROC AUC Score: 0.9800829342343919
Accuracy: 0.9892583120204603
Recall: 0.9521410579345088
Precision: 0.9947368421052631
F1 Score: 0.972972972972973
ROC AUC Score: 0.9754286804435061
Accuracy: 0.9848828420256992
Recall: 0.9521410579345088
Precision: 0.9973614775725593
F1 Score: 0.9742268041237113
ROC AUC Score: 0.9755305721637988
Accuracy: 0.9798590130916415
Recall: 0.9521410579345088
Precision: 0.9973614775725593
F1 Score: 0.9742268041237113
ROC AUC Score: 0.9752316027927578
Accuracy: 0.9707379134860051
Recall: 0.9521410579345088
Precision: 0.9895287958115183
F1 Score: 0.9704749679075738
ROC AUC Score: 0.9709291407924473
# 获取特征重要性
feature_importances = pipe.named_steps['classifier'].feature_importances_
print(feature_importances)
# 获取特征名称
feature_names = pipe.named_steps['preprocessor'].get_feature_names_out()
plt.figure(figsize=(10, 15))
bars = plt.barh(feature_names, feature_importances)
# 在条形图上添加贡献度数值
for bar in bars:
width = bar.get_width()
plt.text(width, bar.get_y() + bar.get_height()/2, f'{width:.2f}', ha='left', va='center')
plt.yticks(rotation=0, fontsize=14)
plt.title('Feature importance', fontsize=14)
plt.show()
[0.24211243 0.04533319 0.01520913 0.03047476 0.02584477 0.01798545
0.32125056 0.02199318 0.02361082 0.01493117 0. 0.02847592
0. 0.017366 0.01342116 0.03188443 0.02479946 0.01728985
0.0137465 0. 0.03046666 0. 0.06380459 0. ]
2. SHAP解释
import shap
# 获取分类器
clf = pipe.named_steps['classifier']
# 获取预处理后的特征
X_preprocessed = pipe.named_steps['preprocessor'].transform(X_train)
# 利用SHAP解释XGBoost模型
explainer = shap.TreeExplainer(clf)
shap_values = explainer.shap_values(X_preprocessed)
# 对于其中的某一行进行分析
shap.initjs()
shap.force_plot(explainer.expected_value, shap_values[2], X_preprocessed[2],feature_names=feature_names)
#对于特征的总体分析
shap.summary_plot(shap_values, X_preprocessed,feature_names=feature_names)
# shap.summary_plot(shap_values, X_preprocessed, plot_type="bar")
#将上面画的图表纵轴中的数值换成对应的变量名
shap.summary_plot(shap_values, X_preprocessed, plot_type="bar", feature_names=feature_names)
# 部分依赖图
shap.dependence_plot(3, shap_values, X_preprocessed, interaction_index=None, show=False)
# 对多个变量的交互进行分析
# shap_interaction_values = shap.TreeExplainer(clf).shap_interaction_values(X_preprocessed)
# shap.summary_plot(shap_interaction_values, X_preprocessed, max_display=6)
# shap.dependence_plot(1, shap_values, X_preprocessed, interaction_index=1, show=False)