/Fine-tuning-IFRE

Primary LanguagePython

Fine-tuning - IFRE

Inplementation of "Relative Fine-tuning in Incremental Few-Shot Relation Extraction"

Environments

  • python 3
  • PyTorch 1.7.1
  • transformers 4.6.0

Datasets and Models

You can find the training and validation data here: FewRel 1.0 data. For the test data, you can easily download from FewRel 1.0 competition website: https://competitions.codalab.org/competitions/27980

We release our trained models using BERT and CP as backend models respectively at Google Drive. The file structure as below:

--BERT
    --nodropPrototype-nodropRelation-lr-1e-5
--CP
    --nodropPrototype-nodropRelation-lr-9e-6
    --nodropPrototype-nodropRelation-lr-5e-6

You can reproduce our result in the paper with models in BERT/nodropPrototype-nodropRelation-lr-1e-5 and CP/nodropPrototype-nodropRelation-lr-5e-6. We also provide the trained model with a different learning rate for CP in CP/--nodropPrototype-nodropRelation-lr-9e-6 for extra reference.

Code

Put all data in the data folder, CP pretrained model in the CP_model folder (you can download CP model from https://github.com/thunlp/RE-Context-or-Names/tree/master/pretrain or Google Drive), and then you can simply use three scripts: run_train.sh, run_eval.sh, run_submit.sh for train, evaluation and test.

Train

Set the corresponding parameter values in the script, and then run:

sh run_train.sh

Some explanations of the parameters in the script:

--pretrain_ckpt
	the path for the BERT-base-uncased
--backend_model
	bert or cp, select one backend model

Evaluation

Set the corresponding parameter values in the script, and then run:

sh run_eval.sh

Some explanations of the parameters in the script:

--test_iter
	1000, the evaluation iteration
--load_ckpt
	the path of the trained model

Test

Set the corresponding parameter values in the script, and then run:

sh run_submit.sh

Some explanations of the parameters in the script:

--test_output
	the path to save the prediction file

Results

BERT on FewRel 1.0 1-shot

Novel Base All
InreProtoNetwork 60.15 82.10 71.13
ICA-Proto 63.25 82.56 72.91
Ours 80.40 78.31 79.36

5-shot

Novel Base All
InreProtoNetwork 65.77 84.64 75.21
ICA-Proto 69.49 84.89 77.19
Ours 83.39 82.67 83.03

Acknowledgment Our code is based on the implementations of SimpleFSRE(https://github.com/lylylylylyly/SimpleFSRE)