harvardnlp / pytorch-struct

Fast, general, and tested differentiable structured prediction in PyTorch
http://harvardnlp.github.io/pytorch-struct
MIT License
1.11k stars 93 forks source link

support KL and cross-entropy semiring #79

Closed sustcsonglin closed 4 years ago

sustcsonglin commented 4 years ago

supporting KL and cross-entropy between two CRF distributions. adding KL and cross-entropy semiring,
and add some test case for kl and cross-entropy

srush commented 4 years ago

This is so cool! Thanks for implementing.

We probably should doc these a bit. Expectation semirings are tricky and seeing the implementation should be really helpful.

Who is this btw?

sustcsonglin commented 4 years ago

Yeah, I think cross-entropy and KL divergence semirings are extremely useful,  it is very worthy to doc these a bit.   I am Songlin Yang and I will be a first-grade postgraduate student in ShanghaiTech University in Sep. , I am very interested in structure prediction and computational linguistics. 

------------------ 原始邮件 ------------------ 发件人: "Sasha Rush"<notifications@github.com>; 发送时间: 2020年7月30日(星期四) 晚上11:58 收件人: "harvardnlp/pytorch-struct"<pytorch-struct@noreply.github.com>; 抄送: "isomorphism"<406736650@qq.com>; "Author"<author@noreply.github.com>; 主题: Re: [harvardnlp/pytorch-struct] support KL and cross-entropy semiring (#79)

This is so cool! Thanks for implementing.

We probably should doc these a bit. Expectation semirings are tricky and seeing the implementation should be really helpful.

Who is this btw?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

srush commented 4 years ago

Very nice to meet you. Two very nice contributions in a hard area of structured prediction. Feel free to keep on editing the library or adding new parts.