zoukankan      html  css  js  c++  java
  • 机器学习使用sklearn进行模型训练、预测和评价

    cross_val_score(model_name, x_samples, y_labels, cv=k)

    作用:验证某个模型在某个训练集上的稳定性,输出k个预测精度。

    K折交叉验证(k-fold)

    把初始训练样本分成k份,其中(k-1)份被用作训练集,剩下一份被用作评估集,这样一共可以对分类器做k次训练,并且得到k个训练结果。

    1 from sklearn.model_selection import cross_val_score
    2 clf = sklearn.linear_model.LogisticRegression()
    3 # X:features  y:targets  cv:k
    4 cross_val_score(clf, X, y, cv=5)

    模型的训练、预测和评价

     1 def svm_model():
     2     from sklearn.metrics import accuracy_score
     3     from sklearn.metrics import precision_score, recall_score, f1_score
     4     from sklearn.svm import SVC
     5     # 模型训练
     6     clf = SVC(kernel='linear')
     7     clf.fit(x_train_samples, y_train_labels)
     8     # 模型存储
     9     joblib.dump(clf, './model/svm_mode.pkl')
    10     # 模型评估
    11     predict_labels = clf.predict(x_test_samples)
    12     Accuracy = accuracy_score(y_test_labels, predict_labels)
    13     Precision = precision_score(y_test_labels, predict_labels, pos_label=0)
    14     Recall = recall_score(y_test_labels, predict_labels, pos_label=0)
    15     F1_scores = f1_score(y_test_labels, predict_labels, pos_label=0)

    整个过程结束。需要说明的是调用K折交叉验证,结果输出的是准确率,其它的指标不会输出。所以,建议还是前期,使用train_test_split()函数划分训练集和验证集,后期根据实际需求评估模型

  • 相关阅读:
    第一课基础知识
    Linux基础命令
    IO&Process基础知识
    caffe-windows 运行matlab 接口的例子
    process 3d image using caffe
    caffe-windows配置 cuda6.5+vs2012
    cuda7.0安装windows+vs2012
    SGD步长
    Exercise: Convolutional Neural NetworkCNN的反向求导及练习
    Create sparse matrix:sparse
  • 原文地址:https://www.cnblogs.com/demo-deng/p/10154222.html
Copyright © 2011-2022 走看看