Paddle crnn的v3 模型训练到什么时候就达到极限准确度?怎么判断?

nx7onnlm  于 2个月前  发布在  其他
关注(0)|答案(9)|浏览(44)

Thank you for contributing to PaddlePaddle. Submitting an issue is a great help for us.
Both Chinese and English issues are welcome.

It's hard to solve a problem when important details are missing.
Before submitting the issue, look over the following criteria before handing your request in.

  • Was there a similar issue submitted or resolved before ? You could search issue in the github.

  • Did you retrieve your issue from widespread search engines ?

  • Is my description of the issue clear enough to reproduce this problem?

  • If some errors occurred, we need details about how do you run your code? , what system do you use? , Are you using GPU or not? , etc.

  • If you use an recording asciinema to show what you are doing to make it happen, that's awesome! We could help you solve the problem more quickly.

  • Is my description of the issue use the github markdown correctly?

  • Please use the proper markdown syntaxes for styling all forms of writing, e.g, source code, error information, etc.

  • Check out this page to find out much more about markdown.

crnn的v3 模型训练到什么时候就达到极限准确度?怎么判断?

fcy6dtqo

fcy6dtqo1#

您好,我们已经收到了您的问题,会安排技术人员尽快解答您的问题,请耐心等待。请您再次检查是否提供了清晰的问题描述、复现代码、环境&版本、报错信息等。同时,您也可以通过查看 官网API文档常见问题历史IssueAI社区 来寻求解答。祝您生活愉快~

Hi! We've received your issue and please be patient to get responded. We will arrange technicians to answer your questions as soon as possible. Please make sure that you have posted enough message to demo your request. You may also check out the APIFAQGithub Issue and AI community to get the answer.Have a nice day!

tvz2xvvm

tvz2xvvm2#

建议固定epoch或者step数进行一次eval,可参考DEMO链接:https://www.paddlepaddle.org.cn/documentation/docs/zh/practices/cv/pointnet.html#sixunlian

ggazkfy8

ggazkfy83#

当eval的准确率不再提升,就是到极限了?

a8jjtwal

a8jjtwal4#

理论上说是的,在eval的数据集上达到了极限。此外建议您在实践中,还可以保存多个准确率较高的checkpoint,在您真实数据集上进行predict,选取最优的训练好的模型

ftf50wuq

ftf50wuq5#

不知道谷歌是怎么训练的,google vision 准确度高很多

xiozqbni

xiozqbni6#

看您之前的eval结果,才训练到第2个epoch,建议看看多训练一会后的的结果

vlf7wbxs

vlf7wbxs7#

数据集有千万。所以比较慢。之前断过一次。共训练了3个epoch。每次eval,如果结果没有改善,比之前的best结果好,还是保存之前的best模型?
谷歌的是怎么训练的?

fcwjkofz

fcwjkofz8#

你可以看一下谷歌训练了多少个epoch。建议是保存多个最优的

hfsqlsce

hfsqlsce9#

不知道谷歌训练多少数据,多少epoch。而且没有谷歌的离线模型可以下载,不知道怎么下载

相关问题