本文共 2255 字，大约阅读时间需要 7 分钟。

在做模型训练的时候，尤其是在训练集上做交叉验证，通常想要将模型保存下来，然后放到独立的测试集上测试，下面介绍的是Python中训练模型的保存和再使用。

 

模型保存（pickle模块和joblib模块）

使用pickle模块或者sklearn内部的joblib

一、使用pickle模块

    from sklearn import svm

    from sklearn import datasets     clf=svm.SVC()     iris=datasets.load_iris()     X,y=iris.data,iris.target     clf.fit(X,y)           import pickle     s=pickle.dumps(clf)     f=open('svm.txt','w')     f.write(s)     f.close()     f2=open('svm.txt','r')     s2=f2.read()     clf2=pickle.loads(s2)     clf2.score(X,y)

二、使用joblib

joblib更适合大数据量的模型，且只能往硬盘存储，不能往字符串存储

    from sklearn.externals import joblib

    joblib.dump(clf,'filename.pkl')     clf=joblib.load('filename.pkl')

三、具体例子如下：

scikit-learn已经有了模型持久化的操作，导入joblib即可

from sklearn.externals import joblib

模型保存

通过joblib的dump可以将模型保存到本地，clf是训练的分类器

    from sklearn.linear_model import LogisticRegression

    from sklearn.svm import SVC     from sklearn.externals import joblib     def test_save_model(self):         model_save_path = "./model_save/"         train_X = [[0, 0], [1, 1]]         train_y = [0, 1]         print "Start LR method ..."         print('LR Train classifier...')         clf = LogisticRegression()         clf.fit(train_X, train_y)         print "LR Model save..."         save_path_name = model_save_path + "lr_" + "train_model.m"         self.is_exist(model_save_path, save_path_name)         joblib.dump(clf, save_path_name)         clf = joblib.load(save_path_name)         print('LR Predict...')         pred = clf.predict_proba(train_X)         submit_csv_name = model_save_path + "lr" + '_submission.csv'         self.make_submission(pred[:, 0], submit_csv_name)               print "Start SVM method ..."         # 训练         print('SVM Train classifier...')         from sklearn import svm         clf = svm.SVC()         clf.fit(train_X, train_y)         # 保存         print "SVM Model save..."         save_path_name=model_save_path+"svm_"+"train_model.m"         self.is_exist(model_save_path,save_path_name)         joblib.dump(clf, save_path_name)         clf = joblib.load(save_path_name)         # 预测         print('SVM Predict...')         pred=clf.predict(train_X)         submit_csv_name = model_save_path+"svm" + '_submission.csv'         self.make_submission(pred, submit_csv_name)               train_X = [[0, 1], [1, 1]]         train_y = [0, 1]         print clf.score(train_X, train_y, sample_weight=None)

模型从本地调回

clf = joblib.load("train_model.m")

通过joblib的load方法，加载保存的模型。

然后就可以在测试集上测试了

clf.predict(test_X) #此处test_X为特征集

———————————————— 版权声明：本文为CSDN博主「Data_IT_Farmer」的原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接及本声明。 原文链接：https://blog.csdn.net/helloxiaozhe/article/details/80658438