yale-lily / convosumm Goto Github PK

Thanks for the awesome work!!! I have several questions about the email dataset in data_processed/email/vanilla

1. what's the difference between email.newsumms and the normal data (such as train.source)
2. Does email.all.newsumms contain email.test.newsumms and email.val.newsumms? And if so, could you please provide a clean email.train.newsumms without test & valid set included?

Hi~
Thank you very much for your code.
But when I tested meeting summarization, I couldn't achieve the results in your paper.Can you provide the hyperparameters?Thanks!

Hi~ thanks for your awesome work, but i met a problem about the results.
When I used your code,I found that my results was lower than you. I don't know that which setting was wrong.
The main hyper-parameters and results are as follows:

The other settings are the same as yours.
I am looking forward to your reply!

Hi , I am trying to replicate your project.As a first step I am trying to download model checkpoints using awscli. I am using aws for the first time and I am having an error " fatal error: An error occurred (AccessDenied) when calling the ListObjectsV2 operation: Access Denied " . Can you help me how to solve this?I created a Free tier account with aws.
Is the model checkpoints are available to download using google drive link also?

I am trying to use scripts/prep.sh and scripts/inference.py to load /reddit_vanilla_actual/checkpoint_best.pt BART for inference. I have been having many issues, mostly related to package versions and the extended 2048 source positions.

Environment:

pytorch                   1.7.1           py3.8_cuda10.2.89_cudnn7.6.5_0    pytorch

And I tried installing fairseq from source to access the examples module, but then I saw you had your own copy of fairseq in this repo so I installed your version according to the instructions here

cd fairseq
pip install --editable ./
python setup.py build develop

I binarized val.source and val.target from and am running inference as such:

python scripts/inference.py /home/aadelucia/ConvoSumm/checkpoints/reddit_vanilla_actual checkpoint_best.pt /home/aadelucia/ConvoSumm/alexandra_test/data_processed /home/aadelucia/ConvoSumm/alexandra_test/data/val.source /home/aadelucia/ConvoSumm/alexandra_test/inference_output.txt 4 1 80 120 1 2048 ./misc/encoder.json ./misc/vocab.bpe

And I get the following error:

Traceback (most recent call last):
  File "scripts/inference.py", line 42, in <module>
    hypotheses_batch = bart.sample(slines, beam=beam, lenpen=lenpen, min_len=min_len, no_repeat_ngram_size=3)
  File "/home/aadelucia/ConvoSumm/code/fairseq/fairseq/hub_utils.py", line 132, in sample
    batched_hypos = self.generate(tokenized_sentences, beam, verbose, **kwargs)
  File "/home/aadelucia/ConvoSumm/code/fairseq/fairseq/models/bart/hub_interface.py", line 108, in generate
    return super().generate(
  File "/home/aadelucia/ConvoSumm/code/fairseq/fairseq/hub_utils.py", line 171, in generate
    for batch in self._build_batches(tokenized_sentences, skip_invalid_size_inputs):
  File "/home/aadelucia/ConvoSumm/code/fairseq/fairseq/hub_utils.py", line 258, in _build_batches
    batch_iterator = self.task.get_batch_iterator(
  File "/home/aadelucia/ConvoSumm/code/fairseq/fairseq/tasks/fairseq_task.py", line 244, in get_batch_iterator
    batch_sampler = dataset.batch_by_size(
  File "/home/aadelucia/ConvoSumm/code/fairseq/fairseq/data/fairseq_dataset.py", line 145, in batch_by_size
    return data_utils.batch_by_size(
  File "/home/aadelucia/ConvoSumm/code/fairseq/fairseq/data/data_utils.py", line 337, in batch_by_size
    return batch_by_size_vec(
  File "fairseq/data/data_utils_fast.pyx", line 20, in fairseq.data.data_utils_fast.batch_by_size_vec
  File "fairseq/data/data_utils_fast.pyx", line 27, in fairseq.data.data_utils_fast.batch_by_size_vec
AssertionError: Sentences lengths should not exceed max_tokens=1024

Am I using the wrong version of a package? Is there something extra needed for this to work?

Hi,

By using the run.sh script in inference mode by loading one of the pretrained models from the README file, I keep getting this error:

Traceback (most recent call last):
  File "./scripts/summarization.py", line 432, in <module>
    main(args)
  File "./scripts/summarization.py", line 346, in main
    model = Summarizer.load_from_checkpoint(args.from_pretrained, args)
  File "/users/oncescu/miniconda3/envs/longformer/lib/python3.7/site-packages/pytorch_lightning/core/saving.py", line 169, in load_from_checkpoint
    model = cls._load_model_state(checkpoint, *args, **kwargs)
  File "/users/oncescu/miniconda3/envs/longformer/lib/python3.7/site-packages/pytorch_lightning/core/saving.py", line 208, in _load_model_state
    model = cls(*cls_args, **cls_kwargs)
TypeError: __init__() takes 2 positional arguments but 3 were given

Hi Alexander,

Thanks for publishing the dataset. After downloading the data, do you know how to split the data to train/valid/test for a fair comparison with your experimental results?

Tyson

Hi,
Thanks for your nice work on conversation summarization. I am planning to use to Argument classifier module for my own work. I have two questions regarding that:

1. I run (arg_classifier_test.py) the argument classifier to label sentences as claim, premise, or non-argumentative units. The softmax prediction also returns predictions for all three classes. Please describe the exact mapping between the softmax prediction vector and the three classes. For example, does the first value in the softmax output represent the claim, premise or a non-argumentative unit?
2. Given a sentence i, it can easily play the role of a claim, premise or non-argumentative unit depending on the topic in discussion. For example, a sentence like "legalizing abortion is equivalent to giving permission to kill a human", can be a premise in the context of an argument about "legalization of abortion" but a non-argumentative unit in the context of an argument about "climate change". However, the current module only takes a sentence as input and not the topic of discussion or any such related context. Do you think this can negatively impact the performance of the argument classifier?