如何使用for循环或条件在pandas数据框的子集中创建多个回归模型（statsmodel）？

Posted 2021-03-28

如何使用for循环或条件在pandas数据框的子集中创建多个回归模型（statsmodel）？

我有一个数据框，其中包含一个具有51个唯一值的变量状态。我必须为每个州制作一个模型。出于某种原因，我仅限于回归（statsmodel），假设变量V1由变量X1，X2，X3预测

状态为1:51，将用作分割该数据帧的条件

如何使用for循环自动执行此任务？

答案

假设您只关注循环而不是将数据帧拆分为51个子部分，这是我尝试您的问题：

可以说，您将OLS功能定义为：

def OLSfunction(y):

    y_train = traindf[y]
    y_test = testdf[y]
    from statsmodels.api import OLS
    x_train = x_traindf
    x_test = x_testdf
    model = OLS(y_train, x_train)
    result = model.fit()
    print (result.summary())
    pred_OLS = result.predict(x_test)
    print("R2", r2_score(y_test, pred_OLS))



Y_s = ['1','2',.....'51']
for y in Y_s:
    y=y
    OLSfunction(y)

请注意，您必须为您要构建模型的特定Y适当地派生traindf和testdf。这些必须正确传递到OLS功能。由于我对数据的外观没有任何看法，所以我没有进入traindf / testdf的拆分/创建......

另一答案

import pandas as pd
import os as os
import numpy as np
import statsmodels.formula.api as sm

首先，我创建了一个dict来保存51个数据集

d = {}
for x in range(0, 52):
    d[x]=ccf.loc[ccf['state'] == x]

d.keys()

dict_keys([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51])

去检查

d[1].head()

然后我在循环中使用dict中的位置运行代码

results={}
for x in range(1, 51):
        results[x] = sm.Logit(d[x].fraudRisk, d[x][names]).fit().summary2()

但是我觉得我应该在sklearn中使用多个分类器。首先，我需要如上所述拆分数据。

from sklearn.model_selection import train_test_split

# Multiple Models
from sklearn.linear_model import LogisticRegression
from sklearn.svm import LinearSVC
from sklearn.neighbors import KNeighborsClassifier  
from sklearn.ensemble import RandomForestClassifier 
from sklearn.naive_bayes import GaussianNB

#Model Metrics
from sklearn.metrics import confusion_matrix
from sklearn.metrics import accuracy_score
from sklearn.metrics import classification_report
from sklearn.metrics import roc_auc_score


lr={}
gnb={}
svc={}
rfc={}
classifier={}
regr_1={}
regr_2={}
import datetime 
datetime.datetime.now() 

for x in range(1, 51):
    X_train, X_test, y_train, y_test = train_test_split(d[x][names], d[x].fraudRisk, test_size=0.3)
    print(len(X_train))
    print(len(y_test))

    # Create classifiers

    lr[x] = LogisticRegression().fit(X_train, y_train).predict(X_test)
    gnb[x] = GaussianNB().fit(X_train, y_train).predict(X_test)
    svc[x] = LinearSVC(C=1.0).fit(X_train, y_train).predict(X_test)
    rfc[x] = RandomForestClassifier(n_estimators=1).fit(X_train, y_train).predict(X_test)
    classifier[x] = KNeighborsClassifier(n_neighbors=3).fit(X_train, y_train).predict(X_test)  

    print(datetime.datetime.now())
    print("Accuracy Score for model for  state ",x, 'is  ')

    print('LogisticRegression',accuracy_score(y_test,lr[x]))
    print('GaussianNB',accuracy_score(y_test,gnb[x]))
    print('LinearSVC',accuracy_score(y_test,svc[x]))
    print('RandomForestClassifier',accuracy_score(y_test,rfc[x]))
    print('KNeighborsClassifier',accuracy_score(y_test,classifier[x]))

    print("Classification Report for model for state ",x, 'is  ')

    print('LogisticRegression',classification_report(y_test,lr[x]))
    print('GaussianNB',classification_report(y_test,gnb[x]))
    print('LinearSVC',classification_report(y_test,svc[x]))
    print('RandomForestClassifier',classification_report(y_test,rfc[x]))
    print('KNeighborsClassifier',classification_report(y_test,classifier[x]))

    print("Confusion Matrix Report for model for state ",x, 'is  ')  

    print('LogisticRegression',confusion_matrix(y_test,lr[x]))
    print('GaussianNB',confusion_matrix(y_test,gnb[x]))
    print('LinearSVC',confusion_matrix(y_test,svc[x]))
    print('RandomForestClassifier',confusion_matrix(y_test,rfc[x]))
    print('KNeighborsClassifier',confusion_matrix(y_test,classifier[x]))

    print("Area Under Curve for model for state ",x, 'is  ') 

    print('LogisticRegression',roc_auc_score(y_test,lr[x]))
    print('GaussianNB',roc_auc_score(y_test,gnb[x]))
    print('LinearSVC',roc_auc_score(y_test,svc[x]))
    print('RandomForestClassifier',roc_auc_score(y_test,rfc[x]))
    print('KNeighborsClassifier',roc_auc_score(y_test,classifier[x]))

花了很长时间为5个型号X 51州提供多个指标，但是值得。让我知道是否有更快或更好的方法来编写更优雅，更少hacky的代码