Die offizielle Implementierung von „Code Synonyms Do Matter: Multiple Synonyms Matching Network for Automatic ICD Coding“ [ACL 2022]
Alle Codes werden unter Python 3.7, PyTorch 1.7.0 getestet. Für Einsum-Berechnungen muss opt_einsum installiert werden. Für das Training der vollständigen MIMIC-III-Einstellung sind mindestens 32 GB GPU erforderlich.
Wir stellen für jeden Datensatz nur mehrere Stichproben bereit. Zum Herunterladen des MIMIC-III-Datensatzes sind Lizenzen erforderlich. Sobald Sie den MIMIC-III-Datensatz erhalten haben, folgen Sie bitte caml-mimic, um den Datensatz vorzuverarbeiten. Nach der Vorverarbeitung sollten Sie train_full.csv , test_full.csv , dev_full.csv , train_50.csv , test_50.csv und dev_50.csv erhalten. Bitte legen Sie sie unter sample_data/mimic3 ab . Dann sollten Sie preprocess/generate_data_new.ipynb zum Generieren eines Datensatzes im JSON-Format verwenden.
Bitte laden Sie word2vec_sg0_100.model von LAAT herunter. Sie müssen den Pfad der Worteinbettung ändern.
MIMIC-III Voll (1 GPU):
CUDA_VISIBLE_DEVICES=0 python main.py --n_gpu 1 --version mimic3 --combiner lstm --rnn_dim 256 --num_layers 2 --decoder MultiLabelMultiHeadLAATV2 --attention_head 4 --attention_dim 512 --learning_rate 5e-4 --train_epoch 20 --batch_size 2 --gradient_accumulation_steps 8 --xavier --main_code_loss_weight 0.0 --rdrop_alpha 5.0 --est_cls 1 --term_count 4 --sort_method random --word_embedding_path word_embedding_path
MIMIC-III Full (8 GPUs):
NCCL_IB_DISABLE=1 CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --nproc_per_node 8 --master_port=1212 --use_env main.py --n_gpu 8 --version mimic3 --combiner lstm --rnn_dim 256 --num_layers 2 --decoder MultiLabelMultiHeadLAATV2 --attention_head 4 --attention_dim 512 --learning_rate 5e-4 --train_epoch 20 --batch_size 2 --gradient_accumulation_steps 1 --xavier --main_code_loss_weight 0.0 --rdrop_alpha 5.0 --est_cls 1 --term_count 4 --sort_method random --word_embedding_path word_embedding_path
MIMIC-III 50:
CUDA_VISIBLE_DEVICES=0 python main.py --version mimic3-50 --combiner lstm --rnn_dim 512 --num_layers 1 --decoder MultiLabelMultiHeadLAATV2 --attention_head 8 --attention_dim 512 --learning_rate 5e-4 --train_epoch 20 --batch_size 16 --gradient_accumulation_steps 1 --xavier --main_code_loss_weight 0.0 --rdrop_alpha 5.0 --est_cls 1 --term_count 8 --word_embedding_path word_embedding_path
python eval_model.py MODEL_CHECKPOINT
mimic3-Kontrollpunkt
mimic3-50 Kontrollpunkt
@inproceedings{yuan-etal-2022-code,
title = "Code Synonyms Do Matter: Multiple Synonyms Matching Network for Automatic {ICD} Coding",
author = "Yuan, Zheng and
Tan, Chuanqi and
Huang, Songfang",
booktitle = "Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)",
month = may,
year = "2022",
address = "Dublin, Ireland",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2022.acl-short.91",
pages = "808--814",
abstract = "Automatic ICD coding is defined as assigning disease codes to electronic medical records (EMRs).Existing methods usually apply label attention with code representations to match related text snippets.Unlike these works that model the label with the code hierarchy or description, we argue that the code synonyms can provide more comprehensive knowledge based on the observation that the code expressions in EMRs vary from their descriptions in ICD. By aligning codes to concepts in UMLS, we collect synonyms of every code. Then, we propose a multiple synonyms matching network to leverage synonyms for better code representation learning, and finally help the code classification. Experiments on the MIMIC-III dataset show that our proposed method outperforms previous state-of-the-art methods.",
}