liming-vie / ruber Goto Github PK

First of all, thanks for your open-source code of this work.
I haven't run the scripts on a particular data set, but I'm curious about the performance that you test on.
Can you reproduce the results of the experiment in the paper?
Thanks!

Could you help share me the data set?? Email:[email protected]

The setting is like this:

1. I train RUBER so that it can evaluate my model, say M.
2. I train M so that it can be evaluated by RUBER.
3. I have a dataset spilt in train, test and valid subsets.

My question is: should the training sets of RUBER and M be different?
Should M be trained on the train set while RUBER be trained on test set?
Thanks!

Hi, first of all, thanks for your open source code of this project.
Inspired by your work, I also reproduce the RUBER and the performance is good.

Here is the link address RUBER.
I use the chinese chit-chat corpus collected by this repo Corpus.

Current, I only provide the code and data. I will show more information and tutorials later.

And thanks for commenting my issue!

There are two undefined names in unreferenced_metric.py, namely reply_sizes and query_sizes. How to fix them? Thanks.

There is a vast number of embeddings nowadays from which I can choose. Which one did you use? Or does it matter?

After reading your code, I have found that the 187 line in unreferenced _metric.py may be a bug, I think that you should delete the not statement.

Do you think that I'm correct?
Please response to me, thanks !