yuxixie / sg-deep-question-generation Goto Github PK

Hello!

Could you please provide the evaluation script?

Regards,
Yokesh

Hello,

Could you suggest the required memory requirements?
When running the preprocessing script, the program crashes at loading Glove vectors. I am running it on a GCP instance with 50 GB of RAM.

Also, could you advise on how to use this model on custom documents?

Regards,
Yokesh

the dataset in google cloud can't open

When I run scripts/preprocess_data.sh, I get the following error:

No such file or directory: '/home/visionx/xj/SG-Deep-Question-Generation-master/datasets/json-data/train.tag.json'

I have downloaded the directory datasets/json-data that has not the file train.tag.json. Could you help me?

Thank you for your great work.
When I run the command line "sh scripts/train_generator.sh", the training process can run in the first three epochs(epoch 0, epoch 1, epoch 2). But when epoch 3, I get the following error:

RuntimeError: "index_select_out_cuda_impl" not implemented for 'Float'.

Could you tell me how to solve this error?

Hi! Thanks for the implementation. I couldn't find in the code where the answer embeddings are averaged and initialized as hidden layer for decoder as mentioned in the paper. I checked the Models.py, Encoders.py and Decoders.py. Also, how are you selecting the answer node out of all the nodes tagged with "ans" attribute as 1?

Hello!
I found when you evaluate the model in the validData, the input of the decoder still includes the question (i.e., the label) if I'm not wrong. I wanna know whether we can input the question of validData (label) in this situation.
Looking forward to your reply, thanks!

the advice allennlp==1.0.0, but when I run the code, give an error:
"allennlp.common.checks.ConfigurationError: key "token_embedders" is required at location "model.text_field_embedder.""
i turn the
allennlp==0.9.0,
spacy==2.1.9,
and install en-core-web-sm 2.2.5 by run python -m spacy download en_core_web_sm,
the allnnlp turn it's code of 1.0.0, this is just for advice

Hi, I'm using allennlp 1.0.0 and allennlp-models 1.0.0.
While running the file get_coref_and_dep_data.py using the command - python preprocess/get_coref_and_dep_data.pn.json data.valid.json dp.json crf_rsltn.json, The following error shows:

File "/-------/------/.local/lib/python3.8/site-packages/allennlp/common/params.py", line 237, in pop value = self.params.pop(key) KeyError: 'token_embedders'
allennlp.common.checks.ConfigurationError: key "token_embedders" is required at location "model.text_field_embedder."
How do you resolve this error? I have tried running the code using 2.7.0 version of allennlp and allennlp-models, but the error remains.

Hello, thanks for your work and patience.
I wonder that how long time I need when train the classifier and generator using one GPU, I found maybe it's a little long when I try to run the code. One week for classifier and one week for generator, two weeks totally. Is there any advice to speed up the training process?
And when I adjust the argument of -gpus, (e.g. 0 1), I found the training process is stopped, has this happened when you train?
Thanks a lot again!

hi:
i can't find the table.3's G.T's question in both valid and train dataset? anyone have the same question or i find in the wrong place?

when I process the data, only find the DP formate, can't find SRL format. does the code not published?

when process data in preprocess_raw_data.py of the coreference_resolution in coreference_resolution:
raw = {d[0]: '\t'.join(d[1]) for d in raw}
seems should be:

raw = {d: '\t'.join(raw[d]) for d in raw}

i find someplace error when run muti-card, for example, in EncoderTransformer, the variable length is a list type in EncoderTransformer ,can't be Assign to multiple cards when run on multi-card

Hello, thanks for your great work!
Maybe there is a bug in the code of "merge.py" about line 85 and 86, sys.argv[2] and sys.argv[3] are in the opposite position.
And can you release the code about how to preprocess the raw data to get "crf_rsltn.json" and "dp.json" and the data of "data.json", "crf_rsltn.json" and "dp.json" ?
Thanks again!!

Hello, I downloaded all the datasets and found that I don’t have permission to download the data in Google Cloud Disk.

Hello,
Can anyone suggest the procedure to be followed to get the prediction on the input provided by us using the same trained model. We want to use our paragraph as input.
Regards,
Shravan