“达观杯”文本智能处理挑战赛④：训练模型之 LR 和 SVM

1. 理解逻辑回归和SVM

之前所写的笔记

算法笔记2——逻辑回归

算法笔记3——从凸优化问题角度看SVM

 

2. 逻辑回归实战

lr = LogisticRegression(C=120,dual=True)
lr.fit(x_train,y_train)

y_test = lr.predict(x_test)

 

3. 线性SVM实战

加入网格搜索，代码如下：

params = {'penalty':['l2','l1'],'C':[2.0,3.0,5.0]}
svc = LinearSVC(dual=False)
clf = GridSearchCV(estimator=svc, param_grid=params, scoring = 'f1_macro', n_jobs=1, cv=3, verbose=3)
clf.fit(x_train,y_train)
print('最优参数：{}'.format(clf.best_params_))

y_test = clf.predict(x_test)

结果如下：

模型训练
Fitting 3 folds for each of 6 candidates, totalling 18 fits
[CV] C=2.0, penalty=l2 ...............................................
[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.
[CV] ...... C=2.0, penalty=l2, score=0.7705630453658429, total=10.2min
[Parallel(n_jobs=1)]: Done   1 out of   1 | elapsed: 10.3min remaining:    0.0s
[CV] C=2.0, penalty=l2 ...............................................
[CV] ...... C=2.0, penalty=l2, score=0.7708993095756499, total= 9.7min
[Parallel(n_jobs=1)]: Done   2 out of   2 | elapsed: 20.1min remaining:    0.0s
[CV] C=2.0, penalty=l2 ...............................................
[CV] ...... C=2.0, penalty=l2, score=0.7658830970101473, total= 9.7min
[CV] C=2.0, penalty=l1 ...............................................
[CV] ...... C=2.0, penalty=l1, score=0.7613864241436358, total=11.9min
[CV] C=2.0, penalty=l1 ...............................................
[CV] ...... C=2.0, penalty=l1, score=0.7594095803538317, total=11.8min
[CV] C=2.0, penalty=l1 ...............................................
[CV] ...... C=2.0, penalty=l1, score=0.7544041642010249, total=11.6min
[CV] C=3.0, penalty=l2 ...............................................
[CV] ...... C=3.0, penalty=l2, score=0.7695717152100418, total=11.2min
[CV] C=3.0, penalty=l2 ...............................................
[CV] ....... C=3.0, penalty=l2, score=0.770736739847966, total=11.1min
[CV] C=3.0, penalty=l2 ...............................................
[CV] ...... C=3.0, penalty=l2, score=0.7659710587250825, total=11.3min
[CV] C=3.0, penalty=l1 ...............................................
[CV] ...... C=3.0, penalty=l1, score=0.7582708388619948, total=12.2min
[CV] C=3.0, penalty=l1 ...............................................
[CV] ...... C=3.0, penalty=l1, score=0.7551676096675127, total=12.0min
[CV] C=3.0, penalty=l1 ...............................................
[CV] ...... C=3.0, penalty=l1, score=0.7478424537131836, total=11.8min
[CV] C=5.0, penalty=l2 ...............................................
[CV] ...... C=5.0, penalty=l2, score=0.7690069285162097, total=12.5min
[CV] C=5.0, penalty=l2 ...............................................
[CV] ...... C=5.0, penalty=l2, score=0.7694686804389069, total=13.4min
[CV] C=5.0, penalty=l2 ...............................................
[CV] ...... C=5.0, penalty=l2, score=0.7654598442185017, total=13.4min
[CV] C=5.0, penalty=l1 ...............................................
[CV] ...... C=5.0, penalty=l1, score=0.7478587997762522, total=11.4min
[CV] C=5.0, penalty=l1 ...............................................
[CV] ...... C=5.0, penalty=l1, score=0.7451328071574291, total=11.0min
[CV] C=5.0, penalty=l1 ...............................................
[CV] ...... C=5.0, penalty=l1, score=0.7394154016122579, total=11.2min
[Parallel(n_jobs=1)]: Done  18 out of  18 | elapsed: 208.4min finished
最优参数：{'C': 2.0, 'penalty': 'l2'}