search

buaadk / KJETE

A KNOWLEDGE/DATA ENHANCED METHOD FOR JOINT EVENT AND TEMPORAL RELATION EXTRACTION implementation code
2 stars 0 forks source link

AttributeError: Can't get attribute 'Event' on <module '__main__'> #1

Open benchen4395 opened 2 years ago

benchen4395 commented 2 years ago

I running the step 4 directly as you said the data have been processed in the data. The commond 'python matres/joint_model_addcommonseStandModel_roberta_matres.py' will cause the following resut: """ Traceback (most recent call last): File "matres/joint_model_addcommonseStandModel_roberta_matres.py", line 825, in exec_func(args) File "matres/joint_model_addcommonseStandModel_roberta_matres.py", line 697, in exec_func train_data = EventDataSet(args.data_dir + type_dir, "train") File "matres/joint_model_addcommonseStandModel_roberta_matres.py", line 673, in init self.data = pickle.load(handle) AttributeError: Can't get attribute 'Event' on <module 'main' from 'matres/joint_model_addcommonseStandModel_roberta_matres.py'> """" So i test the pickle load in the seperate terminal as : """ with open('test.pickle', 'rb') as handle: ...: data = pickle.load(handle) ...: data = list(data.values()) ...: handle.close() """ It also shows "AttributeError: Can't get attribute 'Event' on <module 'main'>" So could you please provide the right running scripts. Thanks

TZWwww commented 1 year ago

hi, have you resolve the problem? thx!

benchen4395 commented 1 year ago

No! We fixed it with other method. However, we failed to reproduce the results in paper with other problems. The first and most important key is that no data-augmentation code and processed aug_data.pickle is provided, we can't reproduce the same result with the data augmentation method described in the subsec 2.3. The other key is that we can't get the comparable performances with the pretrained model Roberta. In fact, we get the much lower results of robert-large compared to bert-base/large. It is confusing!!!

TZWwww commented 1 year ago

Besides, this work wants to tackle the joint event and relation extraction problem. But I'm curious why they don't have the pipeline extraction results(they only have results by using gold events to extract relations).