在 OLS 形式的 StatsModels 中，results.summary 显示回归结果的汇总（例如 AIC、BIC、R-squared、...）。
有没有办法在 sklearn.linear_model.ridge 中有这个汇总表？
如果有人指导我，我将不胜感激。谢谢你。
问问题
15428 次
3 回答
16
据我所知，sklearn 中没有类似 R（或 Statsmodels）的汇总表。（请检查这个答案）
相反，如果您需要它，可以使用 statsmodels.regression.linear_model.OLS.fit_regularized类。（L1_wt=0用于岭回归。）
现在，尽管有下面的文档字符串，它似乎还是会model.fit_regularized(~).summary()返回。None但是对象有params，summary()可以以某种方式使用。
返回： RegressionResults 对象，与 . 返回的类型相同
fit。
例子。
样本数据不适用于岭回归，但无论如何我都会尝试。
在。
import numpy as np
import pandas as pd
import statsmodels
import statsmodels.api as sm
import matplotlib.pyplot as plt
statsmodels.__version__
出去。
'0.8.0rc1'
在。
data = sm.datasets.ccard.load()
print "endog: " + data.endog_name
print "exog: " + ', '.join(data.exog_name)
data.exog[:5, :]
出去。
endog: AVGEXP
exog: AGE, INCOME, INCOMESQ, OWNRENT
Out[2]:
array([[ 38. , 4.52 , 20.4304, 1. ],
[ 33. , 2.42 , 5.8564, 0. ],
[ 34. , 4.5 , 20.25 , 1. ],
[ 31. , 2.54 , 6.4516, 0. ],
[ 32. , 9.79 , 95.8441, 1. ]])
在。
y, X = data.endog, data.exog
model = sm.OLS(y, X)
results_fu = model.fit()
print results_fu.summary()
出去。
OLS Regression Results
==============================================================================
Dep. Variable: y R-squared: 0.543
Model: OLS Adj. R-squared: 0.516
Method: Least Squares F-statistic: 20.22
Date: Wed, 19 Oct 2016 Prob (F-statistic): 5.24e-11
Time: 17:22:48 Log-Likelihood: -507.24
No. Observations: 72 AIC: 1022.
Df Residuals: 68 BIC: 1032.
Df Model: 4
Covariance Type: nonrobust
==============================================================================
coef std err t P>|t| [0.025 0.975]
------------------------------------------------------------------------------
x1 -6.8112 4.551 -1.497 0.139 -15.892 2.270
x2 175.8245 63.743 2.758 0.007 48.628 303.021
x3 -9.7235 6.030 -1.613 0.111 -21.756 2.309
x4 54.7496 80.044 0.684 0.496 -104.977 214.476
==============================================================================
Omnibus: 76.325 Durbin-Watson: 1.692
Prob(Omnibus): 0.000 Jarque-Bera (JB): 649.447
Skew: 3.194 Prob(JB): 9.42e-142
Kurtosis: 16.255 Cond. No. 87.5
==============================================================================
Warnings:
[1] Standard Errors assume that the covariance matrix of the errors is correctly specified.
在。
frames = []
for n in np.arange(0, 0.25, 0.05).tolist():
results_fr = model.fit_regularized(L1_wt=0, alpha=n, start_params=results_fu.params)
results_fr_fit = sm.regression.linear_model.OLSResults(model,
results_fr.params,
model.normalized_cov_params)
frames.append(np.append(results_fr.params, results_fr_fit.ssr))
df = pd.DataFrame(frames, columns=data.exog_name + ['ssr*'])
df.index=np.arange(0, 0.25, 0.05).tolist()
df.index.name = 'alpha*'
df.T
出去。
在。
%matplotlib inline
fig, ax = plt.subplots(1, 2, figsize=(14, 4))
ax[0] = df.iloc[:, :-1].plot(ax=ax[0])
ax[0].set_title('Coefficient')
ax[1] = df.iloc[:, -1].plot(ax=ax[1])
ax[1].set_title('SSR')
出去。
在。
results_fr = model.fit_regularized(L1_wt=0, alpha=0.04, start_params=results_fu.params)
final = sm.regression.linear_model.OLSResults(model,
results_fr.params,
model.normalized_cov_params)
print final.summary()
出去。
OLS Regression Results
==============================================================================
Dep. Variable: y R-squared: 0.543
Model: OLS Adj. R-squared: 0.516
Method: Least Squares F-statistic: 20.17
Date: Wed, 19 Oct 2016 Prob (F-statistic): 5.46e-11
Time: 17:22:49 Log-Likelihood: -507.28
No. Observations: 72 AIC: 1023.
Df Residuals: 68 BIC: 1032.
Df Model: 4
Covariance Type: nonrobust
==============================================================================
coef std err t P>|t| [0.025 0.975]
------------------------------------------------------------------------------
x1 -5.6375 4.554 -1.238 0.220 -14.724 3.449
x2 159.1412 63.781 2.495 0.015 31.867 286.415
x3 -8.1360 6.034 -1.348 0.182 -20.176 3.904
x4 44.2597 80.093 0.553 0.582 -115.564 204.083
==============================================================================
Omnibus: 76.819 Durbin-Watson: 1.694
Prob(Omnibus): 0.000 Jarque-Bera (JB): 658.948
Skew: 3.220 Prob(JB): 8.15e-144
Kurtosis: 16.348 Cond. No. 87.5
==============================================================================
Warnings:
[1] Standard Errors assume that the covariance matrix of the errors is correctly specified.
于 2016-10-17T03:21:58.280 回答
1
from statsmodels.tools.tools import pinv_extended
X_train = sm.add_constant(X_train)
model = sm.OLS(y_train, X_train)
# If True, the model is refit using only the variables that have non-zero
# coefficients in the regularized fit. The refitted model is not regularized.
result = model.fit_regularized(
method = 'elastic_net',
alpha = alp,
L1_wt = l1,
start_params = None,
profile_scale = False,
#refit = True,
refit = False,
maxiter = 9000,
zero_tol = 1e-5,
)
pinv_wexog,_ = pinv_extended(model.wexog)
normalized_cov_params = np.dot(pinv_wexog, np.transpose(pinv_wexog))
final = sm.regression.linear_model.OLSResults(model,
result.params,
normalized_cov_params)
#print(final.summary())
x = sm.add_constant(X_test)
#x = X_test
R2 = r2_score(y_test, final.predict(x))
p = final.pvalues
t = final.tvalues
于 2021-05-22T11:00:15.250 回答
0
感谢@Bugee，我创建了以下函数，该函数返回 statsmodels 指标，例如 rsquared 和 rsquared_adj，以及 summary()。
您也可以使用 sklearn 线性模型（LinearRegression、Lasso、Ridge...）和 statsmodels OLS 和正则化 OLS。
from statsmodels.tools.tools import pinv_extended
import statsmodels.api as sm
import sklearn, statsmodels
def regression_analysis(X, y, model):
is_statsmodels = False
is_sklearn = False
# check for accepted linear models
if type(model) in [sklearn.linear_model._base.LinearRegression,
sklearn.linear_model._ridge.Ridge,
sklearn.linear_model._ridge.RidgeCV,
sklearn.linear_model._coordinate_descent.Lasso,
sklearn.linear_model._coordinate_descent.LassoCV,
sklearn.linear_model._coordinate_descent.ElasticNet,
sklearn.linear_model._coordinate_descent.ElasticNetCV,
]:
is_sklearn = True
elif type(model) in [statsmodels.regression.linear_model.OLS,
statsmodels.base.elastic_net.RegularizedResults,
]:
is_statsmodels = True
else:
print("Only linear models are supported!")
return None
has_intercept = False
if is_statsmodels and all(np.array(X)[:,0]==1):
# statsmodels add_constant has been used already
has_intercept = True
elif is_sklearn and model.intercept_:
has_intercept = True
if is_statsmodels:
# add_constant has been used already
x = X
model_params = model.params
else: # sklearn model
if has_intercept:
x = sm.add_constant(X)
model_params = np.hstack([np.array([model.intercept_]), model.coef_])
else:
x = X
model_params = model.coef_
#y = np.array(y).ravel()
# define the OLS model
olsModel = sm.OLS(y, x)
pinv_wexog,_ = pinv_extended(x)
normalized_cov_params = np.dot(pinv_wexog, np.transpose(pinv_wexog))
return sm.regression.linear_model.OLSResults(olsModel, model_params, normalized_cov_params)
如何使用它？
from sklearn.linear_model import Ridge
skridge = Ridge(alpha=0.2, max_iter=9000, tol=1e-5, fit_intercept=True)
skridge.fit(X,y)
result = regression_analysis(X, y, skridge)
result.summary()
于 2022-01-13T21:21:18.807 回答