/wgc-lstm

A tensorflow implementation of the model proposed in the 2018 ICDM paper titled "A Low Rank Weighted Graph Convolutional Approach to Weather Prediction"

Primary LanguagePythonApache License 2.0Apache-2.0

Overview

A tensorflow implementation of the model proposed in "A Low Rank Weighted Graph Convolutional Approach to Weather Prediction" by Tyler Wilson, Pang-Ning Tan, and Lifeng Luo. Running the demo.py file will train and evaluate a model on the IGRA temperature prediction task described in the paper.

The implementation of the graph convolutional LSTM cell is based on Oliver Hennigh's implementation of a gridded convolutional LSTM cell available here.

When citing, please use: @inproceedings{wilson2018low, title={A Low Rank Weighted Graph Convolutional Approach to Weather Prediction}, author={Wilson, Tyler and Tan, Pang-Ning and Luo, Lifeng}, booktitle={2018 IEEE International Conference on Data Mining (ICDM)}, pages={627--636}, year={2018}, organization={IEEE} }

Installation

To install:

  1. clone the github project
  2. navigate to the cloned project directory on your machine
  3. create a pip virtual environment that uses python 3.5+
  4. activate the pip virtual environment you just created
  5. install the requirements with "pip install -r requirements.txt"
  6. Run the demo with "python demo.py"