知识卡片

普通读书笔记卡 · 7007.g · 生成模型:DeepSeek-V4.1-Flash

渐进式验证

理解等级:高级

对有时间顺序的数据,渐进式验证(也叫回测)更有用:按数据原本出现的顺序喂给模型,每个时间步先让模型对下一条样本做预测并记录进聚合指标,之后才把该样本用于训练——每条数据先评估、后训练。这样能还原真实世界的时间效应,回答“这个模型去年选举日会怎么表现”之类问题,也避免未来信息泄漏。代价有三:模型若需多轮遍历数据才能训好,这套设定就别扭;比较不同模型时必须基于完全相同的时间区间;而且并非所有系统的数据都能像时间那样有公认的规范排序。

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

    • Evaluating Model Validity and Quality · 查看原书依据

      The basic idea is to simulate how training and prediction would work in the real world, but playing the data through to the model in the same order that it originally appeared. For each simulated time step, the model is shown the next example and asked to make its best prediction.

    参考来源

    - 位置:《Reliable Machine Learning》第05章《Evaluating Model Validity and Quality》(源文件:extracted/Reliable Machine Learning/_epub-src/OEBPS/ch05.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:The basic idea is to simulate how training and prediction would work in the real world, but playing the data through to the model in the same order that it originally appeared. For each simulated time step, the model is shown the next example and asked to make its best prediction.