opennre

OpenNRE framework

We adopt OpenNRE project.

This project might be utilized independently from the original one, i.e. installed in a separated folder.

Dependencies

List of the related dependencies:

torch==1.6.0
transformers==3.4.0
numpy==1.19.5
pytest==5.3.2
scikit-learn==0.22.1
scipy==1.4.1
nltk>=3.6.4

The complete pip packages lists experiments were provided.

Usage

Data

You need to provide data in JSONL format at data folder.

By default, we provide a cropped by 100 entries:

• nn-collection (for CNN-based neural networks)
• bert-collection for bert-based models.

Pretrained states

You need to provide the pre-trained state (if you will) at ckpt:

• ra4_DeepPavlov-rubert-base-cased_cls.pth
  • RuAttitudes (4 ep.), with cls based pooling scheme;
• ra4_DeepPavlov-rubert-base-cased_entity.pth
  • RuAttitudes (4 ep.), with entity based pooling scheme;
• ra4-rsr1_DeepPavlov-rubert-base-cased_cls.pth
  • RuAttitudes (4 ep.) + RuSentRel (1 ep.), with cls pooling scheme;
• ra4-rsr1_DeepPavlov-rubert-base-cased_entity.pth
  • RuAttitudes (4 ep.),+ RuSentRel (1 ep.), with entity pooling scheme;
• ra4-rsr1-rsne4_DeepPavlov-rubert-base-cased_cls.pth
  • RuAttitudes (4 ep.) + RuSentRel (1 ep.) + SentiNEREL-train (4 ep.), with cls based pooling scheme;
• ra4-rsr1-rsne4_DeepPavlov-rubert-base-cased_entity.pth
  • RuAttitudes (4 ep.) + RuSentRel (1 ep.) + SentiNEREL-train (4 ep.), with entity based pooling scheme;

Training and Inferring

test_training.py

test_infer.py