helloeve / mre-in-one-pass Goto Github PK

Hi Haoyu,

Nice work on this MRE one-pass! I noticed that the commands in README.md are for SemEval 2018. Do you know what are the commands that you run ACE05 with? (We are especially curious about the hyperparameters.)

Thank you in advance for your information!

Best,
Zhijing

Hi @helloeve ,

If possible, could you please provide the source code for preprocessing the dataset SemEval 2018 Task 7.1? Since the original dataset is in the format XML, I hope you give it for me, as I would like to process the dataset SemEval 2018 Task 7.2. Thanks so much for your consideration.

Hi Haoyu,

In your paper, you reported the performance of the setting "BERT_sp with entity-indicator on input-layer". What is the running command to produce the output of that? Thanks for your help!

Best,
Zhijing

您好：
因为设备的限制，运行您的代码时候，只能设置batch_size=2，并将echo增加，但输出的准确度比论文中的准确度小了20%左右，不知道如何才能复现论文中提出的准确度，所以想来问一下，希望能得到您的答复。
十分感谢

此致
aimasa

Q1: I wonder if train and eval code works ok in run_classifier.py, cuz i encountered error when evaluating which is caused by inconsistent dim of label_ids and predictions.

Q2: why dim of label_ids is [None,max_num_relations] but logits is [None, 6] ?? what does max_num_relations mean?
Could anybody please help me of exact dim transformation in output layer of this model ? Thanks in advance!!

Hi Haoyu,

I noticed that you preprocessed the data files into tsv format. Do you have the preprocessing script for SemEval_raw_data -> your .tsv, and ACE05_data -> your .tsv? We have both data. And if you need, I can show you our LDC license, so that you can feel free to just send me your data. Thank you for making it easier to reproduce your paper and show them in our work following you.

Best,
Zhijing

I am trying to implement just the entity aware self-attention module of the paper and I cannot locate it in the run_classifier.py code. I will be grateful if someone can point me at the self-attention implementation code so I can get my work started.
Thanks in advance.

Error imformation:
INFO:tensorflow:Error recorded from evaluation_loop: Can not squeeze dim[1], expected a dimension of 1, got 12 for 'remove_squeezable_dimensions/Squeeze' (op: 'Squeeze') with input shapes: [?,12].
在调用示例中的多关系抽取时做evaluation时的报错，training时没报错，请问该如何解决？

最近在从事多关系抽取方面的工作，请问能把论文的源代码给我发一份吗，多谢！

Hi Haoyu,

SemEval 2018 Task 7 Subtask 2 is more of Relation Extraction and Task 7 Subtask 1.1 is more of Relation Classification (which does not have "No_Relation"). Is there some reason why you used Subtask 1.1 instead of Subtask 2? Thank you!

Best,
Zhijing

Dear authors,

When we do the following:

!python run_classifier.py --task_name=semeval --do_train=false --do_eval=false --do_predict=true --data_dir=<data-dir> --vocab_file=$BUCKET_NAME/BERT-large-cased/vocab.txt --bert_config_file=$BUCKET_NAME/BERT-large-cased/bert_config.json --init_checkpoint=$BUCKET_NAME/BERT-large-cased/bert_model.ckpt --max_seq_length=128 --predict_batch_size=16 --max_num_relations=12 --do_lower_case=False --use_tpu=True --tpu_name=<address> --output_dir=$OUTPUT_DIR

the following error is displayed

 File "run_classifier.py", line 424, in convert_single_example
    is_real_example=False)
TypeError: __init__() got an unexpected keyword argument 'is_real_example'

My guess is the fix would need to be applied at Line 424? Then the InputFeatures class (Line 178) initialization would work.

Could you kindly provide the list of arguments we would need to supply in order to enable the function? Or at least some direction to resolve the error to run the prediction script on TPU would be most appreciated.

Thank you very much.

Hi,
I have successfully trained the model and did prediction, but I couldn't get the entities along with the relationship predicted. I was wondering which entities from the sentence that the predicted relationship is connecting to. Thanks.