A project for developing transformer-based models for clinical relation extraction

Overview

Clinical Relation Extration with Transformers

Aim

This package is developed for researchers easily to use state-of-the-art transformers models for extracting relations from clinical notes. No prior knowledge of transformers is required. We handle the whole process from data preprocessing to training to prediction.

Dependency

The package is built on top of the Transformers developed by the HuggingFace. We have the requirement.txt to specify the packages required to run the project.

Background

Our training strategy is inspired by the paper: https://arxiv.org/abs/1906.03158 We only support train-dev mode, but you can do 5-fold CV.

Available models

  • BERT
  • XLNet
  • RoBERTa
  • ALBERT
  • DeBERTa
  • Longformer

We will keep adding new models.

usage and example

  • data format

see sample_data dir (train.tsv and test.tsv) for the train and test data format

The sample data is a small subset of the data prepared from the 2018 umass made1.0 challenge corpus

# data format: tsv file with 8 columns:
1. relation_type: adverse
2. sentence_1: ALLERGIES : [s1] Penicillin [e1] .
3. sentence_2: [s2] ALLERGIES [e2] : Penicillin .
4. entity_type_1: Drug
5. entity_type_2: ADE
6. entity_id_1: T1
7. entity_id2: T2
8. file_id: 13_10

note: 
1) the entity between [s1][e1] is the first entity in a relation; the second entity in the relation is inbetween [s2][e2]
2) even the two entities in the same sentenc, we still require to put them separately
3) in the test.tsv, you can set all labels to neg or no_relation or whatever, because we will not use the label anyway
4) We recommend to evaluate the test performance in a separate process based on prediction. (see **post-processing**)
5) We recommend using official evaluation scripts to do evaluation to make sure the results reported are reliable.
  • preprocess data (see the preprocess.ipynb script for more details on usage)

we did not provide a script for training and test data generation

we have a jupyter notebook with preprocessing 2018 n2c2 data as an example

you can follow our example to generate your own dataset

  • special tags

we use 4 special tags to identify two entities in a relation

# the defaults tags we defined in the repo are

EN1_START = "[s1]"
EN1_END = "[e1]"
EN2_START = "[s2]"
EN2_END = "[e2]"

If you need to customize these tags, you can change them in
config.py
  • training

please refer to the wiki page for all details of the parameters flag details

export CUDA_VISIBLE_DEVICES=1
data_dir=./sample_data
nmd=./new_modelzw
pof=./predictions.txt
log=./log.txt

# NOTE: we have more options available, you can check our wiki for more information
python ./src/relation_extraction.py \
		--model_type bert \
		--data_format_mode 0 \
		--classification_scheme 1 \
		--pretrained_model bert-base-uncased \
		--data_dir $data_dir \
		--new_model_dir $nmd \
		--predict_output_file $pof \
		--overwrite_model_dir \
		--seed 13 \
		--max_seq_length 256 \
		--cache_data \
		--do_train \
		--do_lower_case \
		--train_batch_size 4 \
		--eval_batch_size 4 \
		--learning_rate 1e-5 \
		--num_train_epochs 3 \
		--gradient_accumulation_steps 1 \
		--do_warmup \
		--warmup_ratio 0.1 \
		--weight_decay 0 \
		--max_num_checkpoints 1 \
		--log_file $log \
  • prediction
export CUDA_VISIBLE_DEVICES=1
data_dir=./sample_data
nmd=./new_model
pof=./predictions.txt
log=./log.txt

# we have to set data_dir, new_model_dir, model_type, log_file, and eval_batch_size, data_format_mode
python ./src/relation_extraction.py \
		--model_type bert \
		--data_format_mode 0 \
		--classification_scheme 1 \
		--pretrained_model bert-base-uncased \
		--data_dir $data_dir \
		--new_model_dir $nmd \
		--predict_output_file $pof \
		--overwrite_model_dir \
		--seed 13 \
		--max_seq_length 256 \
		--cache_data \
		--do_predict \
		--do_lower_case \
		--eval_batch_size 4 \
		--log_file $log \
  • post-processing (we only support transformation to brat format)
# see --help for more information
data_dir=./sample_data
pof=./predictions.txt

python src/data_processing/post_processing.py \
		--mode mul \
		--predict_result_file $pof \
		--entity_data_dir ./test_data_entity_only \
		--test_data_file ${data_dir}/test.tsv \
		--brat_result_output_dir ./brat_output

Using json file for experiment config instead of commend line

  • to simplify using the package, we support using json file for configuration
  • using json, you can define all parameters in a separate json file instead of input via commend line
  • config_experiment_sample.json is a sample json file you can follow to develop yours
  • to run experiment with json config, you need to follow run_json.sh
export CUDA_VISIBLE_DEVICES=1

python ./src/relation_extraction_json.py \
		--config_json "./config_experiment_sample.json"

Baseline (baseline directory)

  • We also implemented some baselines for relation extraction using machine learning approaches
  • baseline is for comparison only
  • baseline based on SVM
  • features extracted may not optimize for each dataset (cover most commonly used lexical and semantic features)
  • see baseline/run.sh for example

Issues

raise an issue if you have problems.

Citation

please cite our paper:

# We have a preprint at
https://arxiv.org/abs/2107.08957

Clinical Pre-trained Transformer Models

We have a series transformer models pre-trained on MIMIC-III. You can find them here:

Comments
  • prediction on large corpus

    prediction on large corpus

    The package will have issues dealing with the prediction on a large corpus (e.g., thousands of notes). We need to develop a batch process to avoid OOM issue and parallel may be to speed up.

    enhancement 
    opened by bugface 2
  • Not able to get the prediction for Test.csv

    Not able to get the prediction for Test.csv

    Hi

    I am just trying to run the code to get the predictions for the test.csv. i am trying with the pre trained model at https://transformer-models.s3.amazonaws.com/mimiciii_bert_10e_128b.zip.

    While running code I am getting an error as AttributeError: 'BertConfig' object has no attribute 'tags'

    Screen shot of my scree is as below

    image

    opened by vikasgoel2000 1
  • Binary classification with BCELoss or Focal Loss

    Binary classification with BCELoss or Focal Loss

    For binary mode, we currently still use CrossEntropyLoss, but BCELoss is designed for binary classification. We need to add options to use BCELoss or Focal Loss in binary mode

    enhancement 
    opened by bugface 1
  • Ok

    Ok

    Keep forgetting your Singpass username and password? Set it up once on Singpass app for password-free logins next time.

    Download Singpass app at https://app.singpass.gov.sg/share?src=gxe1ax

    opened by Andre11232 0
  • Confused on usage

    Confused on usage

    The input to the prediction model is a .tsv file where the first column is the relation type. So it is unclear to me why we need the model to predict the relation type again.

    Am I misunderstanding? For predicting relations for new data, will the first column be autofilled with NonRel?

    opened by jiwonjoung 1
  • roberta question

    roberta question

    Thank you for providing and actively maintaining this repository. I'm trying to run the roberta on the sample data, but I'm encountering an error (I have tested bert and deberta, and both worked well without any error)

    Here is the code I ran

    export CUDA_VISIBLE_DEVICES=1
    data_dir=./sample_data
    nmd=./roberta_re_model
    pof=./roberta_re_predictions.txt
    log=./roberta_re_log.txt
    
    python ./src/relation_extraction.py \
    		--model_type roberta \
    		--data_format_mode 0 \
    		--classification_scheme 2 \
    		--pretrained_model roberta-base \
    		--data_dir $data_dir \
    		--new_model_dir $nmd \
    		--predict_output_file $pof \
    		--overwrite_model_dir \
    		--seed 13 \
    		--max_seq_length 256 \
    		--cache_data \
    		--do_train \
    		--do_lower_case \
                    --do_predict \
    		--train_batch_size 4 \
    		--eval_batch_size 4 \
    		--learning_rate 1e-5 \
    		--num_train_epochs 3 \
    		--gradient_accumulation_steps 1 \
    		--do_warmup \
    		--warmup_ratio 0.1 \
    		--weight_decay 0 \
    		--max_num_checkpoints 1 \
    		--log_file $log \
    

    but I ran into this error:

    2022-05-12 06:07:50 - Transformer_Relation_Extraction - ERROR - Training error:
    Traceback (most recent call last):
      File "/content/drive/MyDrive/Colab Notebooks/ClinicalTransformer/src/relation_extraction.py", line 59, in app
        task_runner.train()
      File "/content/drive/MyDrive/Colab Notebooks/ClinicalTransformer/src/task.py", line 100, in train
        batch_output = self.model(**batch_input)
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1110, in _call_impl
        return forward_call(*input, **kwargs)
      File "/content/drive/MyDrive/Colab Notebooks/ClinicalTransformer/src/models.py", line 159, in forward
        output_hidden_states=output_hidden_states
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1110, in _call_impl
        return forward_call(*input, **kwargs)
      File "/usr/local/lib/python3.7/dist-packages/transformers/models/roberta/modeling_roberta.py", line 849, in forward
        past_key_values_length=past_key_values_length,
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1110, in _call_impl
        return forward_call(*input, **kwargs)
      File "/usr/local/lib/python3.7/dist-packages/transformers/models/roberta/modeling_roberta.py", line 133, in forward
        token_type_embeddings = self.token_type_embeddings(token_type_ids)
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1110, in _call_impl
        return forward_call(*input, **kwargs)
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/sparse.py", line 160, in forward
        self.norm_type, self.scale_grad_by_freq, self.sparse)
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/functional.py", line 2183, in embedding
        return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
    RuntimeError: Expected tensor for argument #1 'indices' to have one of the following scalar types: Long, Int; but got torch.cuda.FloatTensor instead (while checking arguments for embedding)
    
    Traceback (most recent call last):
      File "/content/drive/MyDrive/Colab Notebooks/ClinicalTransformer/src/relation_extraction.py", line 59, in app
        task_runner.train()
      File "/content/drive/MyDrive/Colab Notebooks/ClinicalTransformer/src/task.py", line 100, in train
        batch_output = self.model(**batch_input)
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1110, in _call_impl
        return forward_call(*input, **kwargs)
      File "/content/drive/MyDrive/Colab Notebooks/ClinicalTransformer/src/models.py", line 159, in forward
        output_hidden_states=output_hidden_states
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1110, in _call_impl
        return forward_call(*input, **kwargs)
      File "/usr/local/lib/python3.7/dist-packages/transformers/models/roberta/modeling_roberta.py", line 849, in forward
        past_key_values_length=past_key_values_length,
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1110, in _call_impl
        return forward_call(*input, **kwargs)
      File "/usr/local/lib/python3.7/dist-packages/transformers/models/roberta/modeling_roberta.py", line 133, in forward
        token_type_embeddings = self.token_type_embeddings(token_type_ids)
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1110, in _call_impl
        return forward_call(*input, **kwargs)
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/sparse.py", line 160, in forward
        self.norm_type, self.scale_grad_by_freq, self.sparse)
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/functional.py", line 2183, in embedding
        return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
    RuntimeError: Expected tensor for argument #1 'indices' to have one of the following scalar types: Long, Int; but got torch.cuda.FloatTensor instead (while checking arguments for embedding)
    Traceback (most recent call last):
      File "/content/drive/MyDrive/Colab Notebooks/ClinicalTransformer/src/relation_extraction.py", line 59, in app
        task_runner.train()
      File "/content/drive/MyDrive/Colab Notebooks/ClinicalTransformer/src/task.py", line 100, in train
        batch_output = self.model(**batch_input)
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1110, in _call_impl
        return forward_call(*input, **kwargs)
      File "/content/drive/MyDrive/Colab Notebooks/ClinicalTransformer/src/models.py", line 159, in forward
        output_hidden_states=output_hidden_states
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1110, in _call_impl
        return forward_call(*input, **kwargs)
      File "/usr/local/lib/python3.7/dist-packages/transformers/models/roberta/modeling_roberta.py", line 849, in forward
        past_key_values_length=past_key_values_length,
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1110, in _call_impl
        return forward_call(*input, **kwargs)
      File "/usr/local/lib/python3.7/dist-packages/transformers/models/roberta/modeling_roberta.py", line 133, in forward
        token_type_embeddings = self.token_type_embeddings(token_type_ids)
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1110, in _call_impl
        return forward_call(*input, **kwargs)
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/sparse.py", line 160, in forward
        self.norm_type, self.scale_grad_by_freq, self.sparse)
      File "/usr/local/lib/python3.7/dist-packages/torch/nn/functional.py", line 2183, in embedding
        return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
    RuntimeError: Expected tensor for argument #1 'indices' to have one of the following scalar types: Long, Int; but got torch.cuda.FloatTensor instead (while checking arguments for embedding)
    
    During handling of the above exception, another exception occurred:
    
    Traceback (most recent call last):
      File "/content/drive/MyDrive/Colab Notebooks/ClinicalTransformer/src/relation_extraction.py", line 181, in <module>
        app(args)
      File "/content/drive/MyDrive/Colab Notebooks/ClinicalTransformer/src/relation_extraction.py", line 63, in app
        raise RuntimeError()
    RuntimeError
    

    Any help would be much appreciated. Thanks for your project!

    opened by jeonge1 4
  • save trained model as a RE model and a core model with only transformer layers

    save trained model as a RE model and a core model with only transformer layers

    we need to separately save the whole RE model and a core transformer model with only transformer layers so that the model can be used for other training tasks.

    enhancement 
    opened by bugface 0
  • ELECTRA and GPT2 support

    ELECTRA and GPT2 support

    Hi,

    I'm wondering how to add ELECTRA and GPT2 support to this module.

    Neither ELECTRA nor GPT2 has pooled output, unlike BERT/RoBERTa-based model.

    I noticed in the models.py the model is implemented as following:

            outputs = self.roberta(
                input_ids,
                attention_mask=attention_mask,
                token_type_ids=token_type_ids,
                position_ids=position_ids,
                head_mask=head_mask,
                output_attentions=output_attentions,
                output_hidden_states=output_hidden_states
            )
    
            pooled_output = outputs[1]
            seq_output = outputs[0]
            logits = self.output2logits(pooled_output, seq_output, input_ids)
    
            return self.calc_loss(logits, outputs, labels)
    

    There are no pooled_output for ELECTRA/GPT2 sequence classification models, only seq_output is in the outputs variable.

    How to get around this limitation and get a working version of ELECTRA/GPT2? Thank you!

    opened by Stochastic-Adventure 2
Releases(v1.0.0)
Owner
uf-hobi-informatics-lab
codebase for hobi informatics lab
uf-hobi-informatics-lab
Malmo Collaborative AI Challenge - Team Pig Catcher

The Malmo Collaborative AI Challenge - Team Pig Catcher Approach The challenge involves 2 agents who can either cooperate or defect. The optimal polic

Kai Arulkumaran 66 Jun 29, 2022
Blender Add-On for slicing meshes with planes

MeshSlicer Blender Add-On for slicing meshes with multiple overlapping planes at once. This is a simple Blender addon to slice a silmple mesh with mul

52 Dec 12, 2022
PyTorch Implementation of Sparse DETR

Sparse DETR By Byungseok Roh*, Jaewoong Shin*, Wuhyun Shin*, and Saehoon Kim at Kakao Brain. (*: Equal contribution) This repository is an official im

Kakao Brain 113 Dec 28, 2022
This repository contains the source code of our work on designing efficient CNNs for computer vision

Efficient networks for Computer Vision This repo contains source code of our work on designing efficient networks for different computer vision tasks:

Sachin Mehta 386 Nov 26, 2022
Code for the paper "How Attentive are Graph Attention Networks?"

How Attentive are Graph Attention Networks? This repository is the official implementation of How Attentive are Graph Attention Networks?. The PyTorch

175 Dec 29, 2022
Implementation of Vaswani, Ashish, et al. "Attention is all you need."

Attention Is All You Need Paper Implementation This is my from-scratch implementation of the original transformer architecture from the following pape

Brando Koch 195 Dec 30, 2022
Datasets, Transforms and Models specific to Computer Vision

torchvision The torchvision package consists of popular datasets, model architectures, and common image transformations for computer vision. Installat

13.1k Jan 02, 2023
SHRIMP: Sparser Random Feature Models via Iterative Magnitude Pruning

SHRIMP: Sparser Random Feature Models via Iterative Magnitude Pruning This repository is the official implementation of "SHRIMP: Sparser Random Featur

Bobby Shi 0 Dec 16, 2021
SketchEdit: Mask-Free Local Image Manipulation with Partial Sketches

SketchEdit: Mask-Free Local Image Manipulation with Partial Sketches [Paper]  [Project Page]  [Interactive Demo]  [Supplementary Material]        Usag

215 Dec 25, 2022
audioLIME: Listenable Explanations Using Source Separation

audioLIME This repository contains the Python package audioLIME, a tool for creating listenable explanations for machine learning models in music info

Institute of Computational Perception 27 Dec 01, 2022
Official PyTorch implementation of Spatial Dependency Networks.

Spatial Dependency Networks: Neural Layers for Improved Generative Image Modeling Đorđe Miladinović   Aleksandar Stanić   Stefan Bauer   Jürgen Schmid

Djordje Miladinovic 34 Jan 19, 2022
Nested Graph Neural Network (NGNN) is a general framework to improve a base GNN's expressive power and performance

Nested Graph Neural Networks About Nested Graph Neural Network (NGNN) is a general framework to improve a base GNN's expressive power and performance.

Muhan Zhang 38 Jan 05, 2023
True per-item rarity for Loot

True-Rarity True per-item rarity for Loot (For Adventurers) and More Loot A.K.A mLoot each out/true_rarity_{item_type}.json file contains probabilitie

Dan R. 3 Jul 26, 2022
EncT5: Fine-tuning T5 Encoder for Non-autoregressive Tasks

EncT5 (Unofficial) Pytorch Implementation of EncT5: Fine-tuning T5 Encoder for Non-autoregressive Tasks About Finetune T5 model for classification & r

Jangwon Park 34 Jan 01, 2023
CoReNet is a technique for joint multi-object 3D reconstruction from a single RGB image.

CoReNet CoReNet is a technique for joint multi-object 3D reconstruction from a single RGB image. It produces coherent reconstructions, where all objec

Google Research 80 Dec 25, 2022
A machine learning benchmark of in-the-wild distribution shifts, with data loaders, evaluators, and default models.

WILDS is a benchmark of in-the-wild distribution shifts spanning diverse data modalities and applications, from tumor identification to wildlife monitoring to poverty mapping.

P-Lambda 437 Dec 30, 2022
A multi-functional library for full-stack Deep Learning. Simplifies Model Building, API development, and Model Deployment.

chitra What is chitra? chitra (चित्र) is a multi-functional library for full-stack Deep Learning. It simplifies Model Building, API development, and M

Aniket Maurya 210 Dec 21, 2022
A novel method to tune language models. Codes and datasets for paper ``GPT understands, too''.

P-tuning A novel method to tune language models. Codes and datasets for paper ``GPT understands, too''. How to use our code We have released the code

THUDM 562 Dec 27, 2022
The Instructed Glacier Model (IGM)

The Instructed Glacier Model (IGM) Overview The Instructed Glacier Model (IGM) simulates the ice dynamics, surface mass balance, and its coupling thro

27 Dec 16, 2022
PyTorch implementation for "Sharpness-aware Quantization for Deep Neural Networks".

Sharpness-aware Quantization for Deep Neural Networks Recent Update 2021.11.23: We release the source code of SAQ. Setup the environments Clone the re

Zhuang AI Group 30 Dec 19, 2022