This repository contains an implementation of paper Automated ICD-9 Coding via A Deep Learning Approach.
All used libraries (Pandas, Numpy, Scikit, Pytorch, and Gensim) are available by default, as part of Google Colab default python3 environment. No additional setup is needed if Google Colab python3 is used as the execution environment.
The Anaconda ennironment.yml file has been provided in the repo.
This environment can be recreated using command:
conda env create -f environment.yml
Please beware that this will install cuda-runtime 11.8.0, along with other supporting cuda packages. So, compatible Nvidia GPU should be present in machine, to leverage GPU during notebook execution.
The training logic for all the models is present in the Notebook dl-model.ipynb. Please follow the instructions in the notebook.
Jupyter Notebook runnable on Google Colab Free Tier: https://drive.google.com/file/d/1i3IcWbIW6hZZL73wg4DmeCjiz5Sa8_as/view?usp=sharing
Necessary pre-processed data files are already available to the above notebook.
The evaluation logic is present in the Notebook pre_trained_models_evaluation.ipynb. Please follow the instructions in the notebook. A dedicated Jupyter Notebook to validate the results for all pre-trained models, runnable on Google Colab Free Tier, is available: https://drive.google.com/file/d/1zKy4eQmnLjnnGDs5Hm_TKdbNMtSPAg_1/view?usp=sharing
Necessary pre-processed data files and pre-trained models are already available to the above notebook.
You can download pretrained models here:
- All pre-trained models (only internal state, and not full models) are available in this Git repo here A dedicated Jupyter Notebook pre_trained_models_evaluation.ipynb is available to build the models from the stored-state, which can be executed in local machine. Same notebook is also available in Google Colab Free Tier: https://drive.google.com/file/d/1zKy4eQmnLjnnGDs5Hm_TKdbNMtSPAg_1/view?usp=sharing