知识卡片
搜索方法没有固定由差到好顺序
网格、随机、进化与贝叶斯搜索适合的空间和预算不同。贝叶斯方法利用已有试验和采集策略选择下一点,不只是最大化代理均值;噪声、并行与维度会影响效率,应与简单基线比较。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
原书出处 · 1 处
- · 查看原书依据
Some common algorithms for hyperparameter tuning include the following (from worst to best): Grid search —Divide the domain of hyperparameters into a discrete grid, and evaluate the model at each point of the grid. Random search —Define the ranges for your hyperparameters, and randomly sample points within these ranges. Evolutionary algorithms —Start with a population of hyperparameter sets; evaluate them; and through an iterative process
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》附录D《附录D:机器学习工具与技术》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/appendix-d.html)
- 结论依据:摘录中关于搜索方法没有固定由差到好顺序的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:Some common algorithms for hyperparameter tuning include the following (from worst to best): Grid search —Divide the domain of hyperparameters into a discrete grid, and evaluate the model at each point of the grid. Random search —Define the ranges for your hyperparameters, and randomly sample points within these ranges. Evolutionary algorithms —Start with a population of hyperparameter sets; evaluate them; and through an iterative process