Threshold function with BP neural network

最新推荐文章于 2022-10-07 22:46:14 发布

Randolph_Lee

最新推荐文章于 2022-10-07 22:46:14 发布

阅读量518

点赞数

分类专栏：机器学习 python 多标签分类文章标签： Python Multi-Label Machine Learning Threshold

本文链接：https://blog.csdn.net/Randolph_Lee/article/details/69372714

版权

本文介绍了在多标签学习中，阈值函数如何用于确定未见实例的正确标签集。通常，通过对每个标签的真实值输出f(x, y)与阈值函数t(x)进行校准来实现这一目标。阈值校准可以通过两种策略完成：常数函数或从训练样例诱导。第一种策略是通过最小化训练集和测试集之间的多标签指示器的差异来设置校准常数，通常选择0.5。第二种策略假设t(x)为线性模型t(x) = w * f(x) + b，并基于训练集解决线性最小二乘问题来求解权重向量w和偏置b。" 113980998,10541559,Yii2数据库表结构缓存优化与清除指南,"['Yii框架', '数据库管理', '缓存技术', 'PHP开发', '性能优化']

摘要由CSDN通过智能技术生成

Threshold function is a common practice in multi-label learning to return as the learned model. In this case, in order to decide the proper label set for unseen instance, the real-valued ouput f(x, y) on each label should be calibrated against the thresholding function output t(x)

Generally, threshold calibration can be accomplished with two strategies. i.e. setting t(x) as constant function or inducing t(x) from the training examples. For the first strategy, the calibration constant can be set to minimize the difference on certain multi-label indicator between the training set and test set. Of course, 0.5 is the popular choice for calibration constant. For the second strategy, the popular is to assume a linear model for t(x): t(x) = w * f(x) + b.To work out the q-dimensional weight vector w and b, the following linear least squares problem is solved based on the training set D.