如何修復 LogisticRegressionCV 中的不收斂

我正在使用 scikit-learn 對一組數據（大約 14 個參數和 > 7000 個標準化觀察值）執行邏輯回歸和交叉驗證。我還有一個目標分類器，其值為 1 或 0。

我遇到的問題是，無論使用哪種求解器，我都會不斷收到收斂警告……

model1 = linear_model.LogisticRegressionCV(cv=10,verbose=1,n_jobs=-1,scoring='roc_auc',solver='newton-cg',penalty='l2')

/home/b/anaconda/lib/python2.7/site-packages/scipy/optimize/linesearch.py:285: LineSearchWarning: The line search algorithm did not converge
 warn('The line search algorithm did not converge', LineSearchWarning)
/home/b/anaconda/lib/python2.7/site-packages/sklearn/utils/optimize.py:193: UserWarning: Line Search failed


model2 = linear_model.LogisticRegressionCV(cv=10,verbose=1,n_jobs=-1,scoring='roc_auc',solver='sag',penalty='l2')

max_iter reached after 2 seconds
max_iter reached after 2 seconds
max_iter reached after 2 seconds
max_iter reached after 2 seconds
max_iter reached after 2 seconds
max_iter reached after 2 seconds
max_iter reached after 2 second

model3 = linear_model.LogisticRegressionCV(cv=10,verbose=1,n_jobs=-1,scoring='roc_auc',solver='lbfgs',penalty='l2')

/home/b/anaconda/lib/python2.7/site-packages/sklearn/linear_model/logistic.py:701: UserWarning: lbfgs failed to converge. Increase the number of iterations.
 warnings.warn("lbfgs failed to converge. Increase the number "

model4 = linear_model.LogisticRegressionCV(cv=10,verbose=1,n_jobs=-1,scoring='roc_auc',solver='liblinear',penalty='l2')

   cg reaches trust region boundary
iter 18 act 1.382e+06 pre 1.213e+06 delta 1.860e+01 f 7.500e+06 |g| 1.696e+06 CG   8
iter  2 act 1.891e+06 pre 1.553e+06 delta 1.060e-01 f 1.397e+07 |g| 1.208e+08 CG   4
iter  4 act 2.757e+04 pre 2.618e+04 delta 1.063e-01 f 1.177e+07 |g| 2.354e+07 CG   4
iter 18 act 1.659e+04 pre 1.597e+04 delta 1.506e+01 f 7.205e+06 |g| 4.078e+06 CG   4
cg reaches trust region boundary
iter  7 act 1.117e+05 pre 1.090e+05 delta 4.146e-01 f 1.161e+07 |g| 9.522e+05 CG   4
iter 31 act 1.748e+03 pre 1.813e+03 delta 2.423e+01 f 6.228e+05 |g| 5.657e+03 CG  14

我需要做什麼才能停止收到警告？

您可以先應用程序的建議來增加max_iter參數；但請記住，您的數據也有可能無法通過邏輯模型擬合。

引用自：https://stats.stackexchange.com/questions/184017