Skip to content

Source code for NeurIPS 2020 paper "Node Classification on Graphs with Few-Shot Novel Labels via Meta Transformed Network Embedding"

License

Notifications You must be signed in to change notification settings

llan-ml/MetaTNE

Repository files navigation

MetaTNE

This repository is the official implementation of MetaTNE in our paper Node Classification on Graphs with Few-Shot Novel Labels via Meta Transformed Network Embedding.

Requirements

We recommend first installing Anaconda3-5.2.0. Then, run the following commands to install requirements:

pip install -U pip && pip uninstall -y numpy && pip install --ignore-installed wrapt numpy==1.17.3 tensorflow-gpu==2.0.0 && pip install networkx==2.2 ray[tune]==0.8.3

To better understand our code, please familiarize yourself with the usage of Ray and Tune.

Usage

You can reproduce the results on BlogCatalog dataset as follows:

  • Data Preparation

    python standardize_data.py --data BlogCatalog
    
  • Training and Evaluation

    python run.py --dataset_str BlogCatalog --meta_num_pos_nodes 10 --meta_num_neg_nodes 20
    

    You may need to modify num_gpus and resources_per_trial according to your computing resources.

  • Result Analysis

    python analysis.py --logdir ~/ray_results/BlogCatalog_10_20
    

About

Source code for NeurIPS 2020 paper "Node Classification on Graphs with Few-Shot Novel Labels via Meta Transformed Network Embedding"

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages