知识卡片

普通读书笔记卡 · 6723.c · 生成模型:GPT-6

挑选幸运种子也会消耗验证信息

理解等级:中级

记录随机种子、数据和软件有助重现一次结果,但反复选最高分的种子属于调参。若对最终测试集这样选择,就会高估选型后的表现;一个win标志不证明找到了彩票假说意义上的稀疏子网。

这张卡片看不懂?

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

      Did you notice the win=True flag in listing 7.30? That’s an Easter egg we created within our CNN pipeline. Whenever we discover a winning ticket in the “lottery ticket hypothesis” game, we hardcode it into our pipeline. For this to work, you have to keep track of the random seeds you use and the exact dataset and software you are using. If you can recreate

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第07章《卷积网络识别文本模式》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-7.html) - 结论依据:摘录中关于挑选幸运种子也会消耗验证信息的定义、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:Did you notice the win=True flag in listing 7.30? That’s an Easter egg we created within our CNN pipeline. Whenever we discover a winning ticket in the “lottery ticket hypothesis” game, we hardcode it into our pipeline. For this to work, you have to keep track of the random seeds you use and the exact dataset and software you are using. If you can recreate