Multi Filter Residual Convolutional Neural Network for Text Classification

The Multi Filter Residual Convolutional Neural Network (MultiResCNN) is built based on TextCNN, Residual Network and CAML. It could be used as a strong baseline model for text classification. The repo can be used to reproduce the results in the paper:

@inproceedings{li2020multirescnn,  
 title={ICD Coding from Clinical Text Using Multi-Filter Residual Convolutional Neural Network},  
 author={Li, Fei and Yu, Hong},  
 booktitle={Proceedings of the Thirty-Fourth AAAI Conference on Artificial Intelligence},
 year={2020}  
}

Setup

This repo mainly requires the following packages.

• gensim 3.4.0
• nltk 3.3
• numpy 1.15.1
• python 3.6.7
• pytorch 1.0.1
• scikit-learn 0.20.0
• allennlp 0.8.4
• pytorch-pretrained-bert 0.6.2

Full packages are listed in requirements.txt.

Usage

1. Prepare data

Our process of preparing data just follows CAML with slight modifications. Put the files of MIMIC III and II into the 'data' dir as below:

data
|   D_ICD_DIAGNOSES.csv
|   D_ICD_PROCEDURES.csv
└───mimic2/
|   |   MIMIC_RAW_DSUMS
|   |   MIMIC_ICD9_mapping
|   |   training_indices.data
|   |   testing_indices.data
└───mimic3/
|   |   NOTEEVENTS.csv
|   |   DIAGNOSES_ICD.csv
|   |   PROCEDURES_ICD.csv
|   |   *_hadm_ids.csv (get from CAML)

Run python preprocess_mimic3.py and python preprocess_mimic2.py.

2. Train and test using full MIMIC-III data

python main.py -data_path ./data/mimic3/train_full.csv -vocab ./data/mimic3/vocab.csv -Y full -model MultiResCNN -embed_file ./data/mimic3/processed_full.embed -criterion prec_at_8 -gpu 0 -tune_wordemb

3. Train and test using top-50 MIMIC-III data

python main.py -data_path ./data/mimic3/train_50.csv -vocab ./data/mimic3/vocab.csv -Y 50 -model MultiResCNN -embed_file ./data/mimic3/processed_full.embed -criterion prec_at_5 -gpu 0 -tune_wordemb

4. Train and test using full MIMIC-II data

python main.py -data_path ./data/mimic2/train.csv -vocab ./data/mimic2/vocab.csv -Y full -version mimic2 -model MultiResCNN -embed_file ./data/mimic2/processed_full.embed -criterion prec_at_8 -gpu 0 -tune_wordemb  

5. If you want to use ELMo, add -use_elmo on the above commands.

6. Train and test using top-50 MIMIC-III data and BERT

python main.py -data_path ./data/mimic3/train_50.csv -vocab ./data/mimic3/vocab.csv -Y 50 -model bert_seq_cls -criterion prec_at_5 -gpu 0 -MAX_LENGTH 512 -bert_dir <your bert dir>

Acknowledgement

We thank all the people that provide their code to help us complete this project.