知识卡片
继续训练可能扩大泛化差距
本章训练继续从7轮到20轮,训练准确率提高而留出表现略下降,提示更多拟合未带来同等泛化收益。可按验证监控、早停和正则选择轮数;这个曲线不是人类能力上界,也不能据此声称模型胜过专家。
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
What happens if you continue the training for a bit longer? Listing 7.21 Continuing the training pipeline.epochs = 13 #1 pipeline = pipeline.train()Epoch: 1, loss: 0.24797, Train accuracy: 0.84528, Test accuracy: 0.78740 Epoch: 2, loss: 0.16067, Train accuracy: 0.86528, Test accuracy: 0.78871 ... Epoch: 12, loss: 0.04796, Train accuracy: 0.93578, Test accuracy: 0.77690 Epoch: 13, loss: 0.13394, Train accuracy: 0.94132, Test accuracy: 0.77690 #1 7
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:摘录中关于继续训练可能扩大泛化差距的定义、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:What happens if you continue the training for a bit longer? Listing 7.21 Continuing the training pipeline.epochs = 13 #1 pipeline = pipeline.train()Epoch: 1, loss: 0.24797, Train accuracy: 0.84528, Test accuracy: 0.78740 Epoch: 2, loss: 0.16067, Train accuracy: 0.86528, Test accuracy: 0.78871 ... Epoch: 12, loss: 0.04796, Train accuracy: 0.93578, Test accuracy: 0.77690 Epoch: 13, loss: 0.13394, Train accuracy: 0.94132, Test accuracy: 0.77690 #1 7