raldir/zero-shot-nerc

Leveraging Type Descriptions for Zero-shot Named Entity Recognition and Classification, published in ACL 2021.

zero-shot-nerc's Issues

请教GPU内存

您好，想请教下您用的8000显卡的内存是多少呢？
我用了2个P100，分别是32寸的，batch size设置为1了，在medmentions数据集上还是会爆显存，是因为内存不足吗？

Missing classes and methods

Hi,
I'm trying to run the test script as described in the documentation, but I notice that some classes/methods are missing in the repository:

• DataloaderSimple: https://github.com/Raldir/Zero-shot-NERC/blob/main/src/test.py#L37
• evaluate_baseline: https://github.com/Raldir/Zero-shot-NERC/blob/main/src/test.py#L43
• EntityClassificationDataset: https://github.com/Raldir/Zero-shot-NERC/blob/main/src/test.py#L47
• EntityClassificationMultiClassDataset: https://github.com/Raldir/Zero-shot-NERC/blob/main/src/test.py#L49
• evaluator_classification: https://github.com/Raldir/Zero-shot-NERC/blob/main/src/models/trainer.py#L376

Please provide best trained models

Can you please provide best trained models.

error KeyError: 'T116,T123' while running generate_datasets.py to preprocess 'MedMentions' data

I am getting error KeyError: 'T116,T123' while running generate_datasets.py to preprocess MedMentions data.
Below is command I ran.

python generate_datasets.py --dataset medmentions --mode preprocess --input_path ./data/MedMentions --output_path /dccstor/dinesh/ZNER_baselines/formatted_data/corpus_pubtator.txt

I have downloaded MedMentions data from https://github.com/chanzuckerberg/MedMentions/tree/master/full/data.

The first mention in corpus_pubtator.txt is 25763772 0 5 DCTN4 T116,T123 C4308010 .

The error is because entity DCTN4 is linked to two semantic types: "Amino Acid, Peptide, or Protein" (T116) and "Biologically Active Substance" (T123). The current code does not handle cases where an entity has more than one semantic types.