知识卡片

普通读书笔记卡 · 7005.c · 生成模型:DeepSeek-V4.1-Flash

留出验证集与过拟合

理解等级:初级

训练出的模型没有「标准答案」可对照,没人能只凭知识判断好坏,最稳妥的是各种压力测试与验证。最基础的做法是从已备好的训练数据里随机留出一部分、称 held-out test set 且不用于训练;因模型从未见过它,可当作「未来未见数据」的替身,把每条样本的预测与真实标签比对来衡量表现。验证数据必须与训练严格隔离,因为模型极易过拟合——把训练数据背得滚瓜烂熟,却无法对新的未见数据做出好预测。

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

    • Basic Introduction to Models · 查看原书依据

      It is critical that this validation data is indeed held out from training, because it is all too easy for a model to experience overfitting, which happens when the model memorizes its training data perfectly but cannot predict well on new unseen data. Once we are happy with the validation of our model, it is time to deploy it in the overall production system.

    参考来源

    - 位置:《Reliable Machine Learning》第03章《Basic Introduction to Models》(源文件:extracted/Reliable Machine Learning/_epub-src/OEBPS/ch03.xhtml) - 结论依据:原文明确阐述该主题,并给出定义、实践或反例。 - 原始内容:It is critical that this validation data is indeed held out from training, because it is all too easy for a model to experience overfitting, which happens when the model memorizes its training data perfectly but cannot predict well on new unseen data. Once we are happy with the validation of our model, it is time to deploy it in the overall production system.