问题描述
对于使用 python 的模块化 statsmodels ，我想知道使用 statsmodels.formula.api 调用同一过程的不同之处与 statsmodels.regression.quantile_regression 相对应.特别是，我获得了参数估计的差异.
For the modul statsmodels using python, I would please like to know how differences in calling the same procedures using statsmodels.formula.api versus statsmodels.regression.quantile_regression come about.In particular, I obtain differences in parameter estimates.
随附了一个最小的工作示例.
A minimum working example is attached.
#%% Moduls;
import numpy as np
import pandas as pd
import statsmodels.api as sm
import statsmodels.formula.api as smf
from statsmodels.regression.quantile_regression import QuantReg
#%% Load in sample data;
data = sm.datasets.engel.load_pandas().data
#%% smf-Version;
model1 = smf.quantreg(formula='foodexp ~ income', data=data, missing="drop")
result1 = model1.fit(q=0.5, vcov='robust', kernel='epa', bandwidth='hsheather', max_iter=1000, p_tol=1e-06)
#%% QuantReg-Version;
model2 = QuantReg \
(
data['foodexp'].values,
exog = sm.tools.tools.add_constant(data['income']).values,
missing = 'drop'
)
result2 = model2.fit \
(
q = 0.5, vcov='robust', kernel='epa', bandwidth='hsheather', max_iter=1000, p_tol=1e-06
)
#%% Compare Results;
print(result1.params[0])
print(result2.params[0])
print('Difference times 10^9: ' + str(abs(10**9*(result1.params[0]-result2.params[0]))))
我需要编辑我的问题；我仍然非常感谢下面提出的解决方法，在应用的环境中不起作用；原因:我没有1个回归变量.请找到附件的修改版本.
I need to edit my question; the workaround proposed by below, for which I am still very grateful, does not work in the applied setting; reason: I do not have only 1 regressor.Please find the modified version attached.
#%% Moduls;
import statsmodels.api as sm
import statsmodels.formula.api as smf
from statsmodels.regression.quantile_regression import QuantReg
#%% Load in sample data;
data = sm.datasets.engel.load_pandas().data
data['income2'] = data['income']**2
#%% smf-Version;
model1 = smf.quantreg(formula='foodexp ~ income + income2', data=data, missing="drop")
result1 = model1.fit(q=0.5, vcov='robust', kernel='epa', bandwidth='hsheather', max_iter=1000, p_tol=1e-06)
#%% QuantReg-Version;
model2 = QuantReg \
(
data['foodexp'].values,
exog = sm.tools.tools.add_constant(data[['income', 'income2']].values),
missing = 'drop'
)
result2 = model2.fit \
(
q = 0.5, vcov='robust', kernel='epa', bandwidth='hsheather', max_iter=1000, p_tol=1e-06
)
#%% Compare Results;
print(result1.params[0])
print(result2.params[0])
print('Difference times 10^9: ' + str(abs(10**9*(result1.params[0]-result2.params[0]))))
推荐答案
您需要对代码进行一些小的更改.这有很大的不同
You need a small change in your code. That's making a big difference
#%% QuantReg-Version;
model2 = QuantReg ( data['foodexp'].values, exog = sm.tools.tools.add_constant(data['income'].values), missing = 'drop')
正如您所说的那样，它在内部实现上有很大的不同.
As you are putting it outside is making a big difference in internal implementation.
最终实施
#%% Moduls;
import numpy as np
import pandas as pd
import statsmodels.api as sm
import statsmodels.formula.api as smf
from statsmodels.regression.quantile_regression import QuantReg
#%% Load in sample data;
data = sm.datasets.engel.load_pandas().data
#%% smf-Version;
model1 = smf.quantreg(formula='foodexp ~ income', data=data, missing="drop")
result1 = model1.fit(q=0.5, vcov='robust', kernel='epa', bandwidth='hsheather',
max_iter=1000, p_tol=1e-06)
#%% QuantReg-Version;
model2 = QuantReg \
(
data['foodexp'].values,
exog = sm.tools.tools.add_constant(data['income'].values),
missing = "drop"
)
result2 = model2.fit(q=0.5, vcov='robust', kernel='epa', bandwidth='hsheather', max_iter=1000, p_tol=1e-06)
#%% Compare Results;
print(result1.params[0])
print(result2.params[0])
print('Difference times 10^9: ' + str(abs(10**9*(result1.params[0]-result2.params[0]))))
除了上面的代码.我已将exog从模型2复制到模型1
Addition to my above code. I have copied exog from model 2 to model 1
#%% Moduls;
import statsmodels.api as sm
import statsmodels.formula.api as smf
from statsmodels.regression.quantile_regression import QuantReg
#%% Load in sample data;
data = sm.datasets.engel.load_pandas().data
data['income2'] = data['income']**2
model1 = smf.quantreg(formula='foodexp ~ income + income2', data=data, missing="drop")
model2 = QuantReg (data['foodexp'].values, exog = sm.tools.tools.add_constant(data[['income', 'income2']].values), missing = 'drop')
model1.exog = model2.exog
result1 = model1.fit(q=0.5, vcov='robust', kernel='epa', bandwidth='hsheather', max_iter=1000, p_tol=1e-06)
result2 = model2.fit(q=0.5, vcov='robust', kernel='epa', bandwidth='hsheather', max_iter=1000, p_tol=1e-06)
#%% Compare Results;
print(result1.params[0])
print(result2.params[0])
print('Difference times 10^9: ' + str(abs(10**9*(result1.params[0]-result2.params[0]))))
第二种方法:-我将exog从模型1复制到模型2
And second approach:- I have copied exog from model 1 to model 2
#%% Moduls;
import statsmodels.api as sm
import statsmodels.formula.api as smf
from statsmodels.regression.quantile_regression import QuantReg
#%% Load in sample data;
data = sm.datasets.engel.load_pandas().data
data['income2'] = data['income']**2
model1 = smf.quantreg(formula='foodexp ~ income + income2', data=data, missing="drop")
model2 = QuantReg (data['foodexp'].values, exog = sm.tools.tools.add_constant(data[['income', 'income2']].values), missing = 'drop')
model2.exog = model1.exog
result1 = model1.fit(q=0.5, vcov='robust', kernel='epa', bandwidth='hsheather', max_iter=1000, p_tol=1e-06)
result2 = model2.fit(q=0.5, vcov='robust', kernel='epa', bandwidth='hsheather', max_iter=1000, p_tol=1e-06)
#%% Compare Results;
print(result1.params[0])
print(result2.params[0])
print('Difference times 10^9: ' + str(abs(10**9*(result1.params[0]-result2.params[0]))))
如果我两个表情都保持相同的值，则答案是相等的.因此，我在前面提到的数据转换的实现方式上存在明显差异.
If i keep both exog to same values, answers are equal. So there is clear difference in implementation for data conversion i stated previously.
这篇关于python statsmodels:输出"formula.api"的差异对比"regression.quantile_regression"的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持！