知识卡片
输入清理需要受控比较收益
错拼修正、收缩展开及空白处理可能改善某些输入,也可能破坏专名、风格和原tokenizer契约。保留original_text并比较可帮助定位变化,仍应固定划分并做消融;与更多训练和长窗口一起改时不能单独归因清理。
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
Building upon the training code you executed in the previous example, now it’s time to work on improving your model’s accuracy. Cleaning the text a bit with some preprocessing is fairly straightforward. The book’s example source code comes with a utility TextPreprocessor class we authored to replace common misspellings, expand contractions, and perform other miscellaneous cleaning, such as removing extra whitespace characters. Go ahead and
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》第09章《可堆叠的Transformer》(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-9.html)
- 结论依据:摘录中关于输入清理需要受控比较收益的定义、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:Building upon the training code you executed in the previous example, now it’s time to work on improving your model’s accuracy. Cleaning the text a bit with some preprocessing is fairly straightforward. The book’s example source code comes with a utility TextPreprocessor class we authored to replace common misspellings, expand contractions, and perform other miscellaneous cleaning, such as removing extra whitespace characters. Go ahead and