Dirk Groeneveld

Results 200 comments of Dirk Groeneveld

I'm not super familiar with that code, but it seems to me that you'll have to modify `LstmCellDecoderNet` to use two `LSTMCell`s instead of just one. You can modify `previous_state`...

I am also interested in getting this done, but without a clear indication of what exactly the format is, I don't see how we can do it.

The extended positional embeddings would be great. I assume it would be relatively self-contained, just resizing the embedding matrix during model initialization?

Is this just about the warning? I think the tokenizers will tokenize just fine, but they want to print that warning. If you know of a way to silence the...

Is this something we could adopt as a metric in AllenNLP? The metric API would require that we pass in two strings (or two lists of strings), one with the...

There is no point in worrying about differences in the implementation of ROUGE while one method uses stemmed tokens, and another uses word pieces. Let's get a version running that...

Oh, I see that https://pypi.org/project/rouge-score/ does all of it, tokenization, stemming, the lot. Then it should be easy to use it in an AllenNLP metric.

@niansong1996, your implementation of `NEW_ROUGE`, it uses the Google package? Can you point me to that code? I want to see how hard it would be to make it an...