Could we use the dev data as a part of our training data in the end?

[Bill Yuchen Lin]

Hi Leon,

I have a question about the use of dev data. I know that dev data is for us to prevent overfitting and tune the hyper parameters. But after that, say I have already decided the hyper-parameters and am going to label the test data for the final submission. Could we use the dev data as a part of our final training data to re-train a final model with the previously decided hyper parameters? Is this allowed? 

It’s my first time to participate in a real shared task, please forgive my ignorance if it’s a common rule that we shouldn’t do that.

Thanks and regards,

Bill

[Leon Derczynski]

Hi,

Sure, we consider the dev data as data to be used however you like. Certainly, we have no way of stopping people using the dev data however they desire.

All the best,

Leon

--
You received this message because you are subscribed to the Google Groups "Workshop on Noisy User-generated Text (WNUT)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to wnut+unsubscribe@googlegroups.com.
To post to this group, send email to wn...@googlegroups.com.
Visit this group at https://groups.google.com/group/wnut.
To view this discussion on the web visit https://groups.google.com/d/msgid/wnut/FE18B701-8A4E-4231-9070-6F4AE8D87CFC%40sjtu.edu.cn.
For more options, visit https://groups.google.com/d/optout.