A PyTorch implementation of GraphSAGE for Medical Ontology embedding

Lui Sheng Jie (luishengjie@outlook.com; luishengjie@u.nus.edu)

Forked from: A PyTorch implementation of GraphSAGE

• This package contains a PyTorch implementation of GraphSAGE.

• This package is a fork of the GraphSage implementation by Tianwen Jiang (tjiang2@nd.edu), Tong Zhao (tzhao2@nd.edu), Daheng Wang (dwang8@nd.edu).

Environment settings

• python==3.6.8
• pytorch==1.0.0

Basic Usage

Main Parameters:

--dataSet     The input graph dataset. (default: cora)
--agg_func    The aggregate function. (default: Mean aggregater)
--epochs      Number of epochs. (default: 50)
--b_sz        Batch size. (default: 20)
--seed        Random seed. (default: 824)
--unsup_loss  The loss function for unsupervised learning. ('margin' or 'normal', default: normal)
--config      Config file. (default: ./src/experiments.conf)
--cuda        Use GPU if declared.

Learning Method

The user can specify a learning method by --learn_method, 'sup' is for supervised learning, 'unsup' is for unsupervised learning, and 'plus_unsup' is for jointly learning the loss of supervised and unsupervised method.

Example Usage

To run the unsupervised model on Cuda:

python -m src.main --epochs 50 --cuda --learn_method unsup

Generate ICD10 Graph

To generate the edgelist for the ICD10 ontology, run the following script:

python src/icd10_graph.py --outdir icd10-data

python -m src.main --epochs 3 --learn_method unsup --dataSet='icd10'