ValueError: Solver lbfgs supports only ‘l2‘ or ‘none‘ penalties, got l1 penalty.

逻辑回归代码段：

model = LogisticRegression(penalty=reg)

在进行逻辑回归预测时报错：

ValueError: Solver lbfgs supports only 'l2' or 'none' penalties, got l1 penalty.

原因：

LogisticRegression(C=0.01, class_weight=None, dual=False, fit_intercept=True,
                   intercept_scaling=1, l1_ratio=None, max_iter=100,
                   multi_class='auto', n_jobs=None, penalty='l1',
                   random_state=None, solver='lbfgs', tol=0.0001, verbose=0,
                   warm_start=False)

问题就出现在solver这个参数上，这个参数定义的是分类器，‘newton-cg’，‘sag’和‘lbfgs’等solvers仅支持‘L2’regularization，‘liblinear’ solver同时支持‘L1’、‘L2’regularization，若dual=Ture，则仅支持L2 penalty。

决定惩罚项选择的有2个参数：dual和solver，如果要选L1范数，dual必须是False，solver必须是liblinear

问题搞清楚了，把上面代码改成：
 

model = LogisticRegression(penalty=reg, solver='liblinear')

即可正常运行：

l1
训练集MSE:0.026373626373626374
测试集MSE:0.07894736842105263
训练测试MSE差异0.05257374204742625
训练集klloss: 0.9109286002545062
测试集klloss 2.7267735609115524
l2
训练集MSE:0.03076923076923077
测试集MSE:0.07894736842105263
训练测试MSE差异0.048178137651821856
训练集klloss: 1.0627491549517625
测试集klloss 2.7267735609115524

Process finished with exit code 0