zhangshiyue / qgforqa Goto Github PK

hello,
i have some questions about the bert_qg vocab. you just given the preprocessed vocab file but the codes to get vocab and related embed are not in the preprocess.py. would you like to release this part of code?

Hi, this work is really impressive! Now I want to use your BERT QG model to train on my dataset. Can you give a link of your pre-trained BERT-QPC and BERT-QA model?

I have a quick question about how BERT Embedding are used in this model. Based on this line it seems that questions are only generated if there is a ground truth to get the CLS token from BERT. Is this what is going on behind the scenes? I'm currently trying to implement a similar model in PyTorch and I'm running into an issue on what to use as the first input to the decoder and it centers on the highly-contextualized nature of BERT.

Thanks,
Matthew.

Loading model...
data/experimental/test.tfrecords
USING SKIP CONNECTIONS
No Best!

I tried $MODEL "model_elmo_qg" but this issue happens ..
What should I do more ??

Hi, thanks for the very nice work.
I wonder if you will release the code for RL model?

Hi, I totally follow your step and train the ELMO-QG baseline model but I only got 13.9 BLEU-4 scores. It seems that something was wrong somewhere ....

How to generate the questions from the text I have,
Processing etc?

Hi, thank you very much for publishing the codes and data.
They are very helpful because few research papers in QG publish their code or data.

Now I am trying to reproduce the result of the semi-supervised QA result in your paper using HarvestingQA dataset (https://github.com/xinyadu/harvestingQA/tree/master/dataset).

However, when I use your preprocess.py to preprocess HarvestingQA dataset, I got many errors in:

https://github.com/ZhangShiyue/QGforQA/blob/master/QG/ELMo_QG/prepare.py#L54

In total, 1189314 examples in HarvestingQA got this error here.
I used the train.json of HarvestingQA.
How did you cope with this in your experiment?
I would appreciate it if you could show the preprocessing procedure in your semi-supervised QA experiment.
Thank you very much.

Hi @ZhangShiyue ,
I had downloaded all data sets need for the project according to README.md file can you tell me the next steps to be done fro training and testing usage for generating of questions and answers...

Thanks and Regards,
Manikantha Sekhar.

Hi,

Thank you for sharing this awesome code! I noticed that you also reported Q-BLEU1 score in your paper. I was wondering if you could kindly share the script for computing Q-BLEU1 score as well. I had some trouble using the official repo to compute Q-BLEU1, as detailed here.

Thank you!

hello, do you have a code version about PyTorch to share?

Hi Shiyue,

Thanks for the nice job.
I realized from your code that for each checkpoint you are saving the best model based on BLEU score on validation set. Is that right?
Just wanted double check what is the metric you used to save the best model? BLEU or your own reward scores?

Thanks