GCN-LSTM

Multivariate Time Series-based Solar Flare Prediction by Functional Network Embedding and Sequence Modeling

Figure: GCN-based node-attributed functional network embedding and LSTM-based local and global sequence embedding

Overview

This repository provides the implementation of MVTS-based Solar Flare Prediction using GCN and LSTM. The repository is organised as follows:

• data/ contains the necessary datasets.
• models/ contains the implementation of the GCN-LSTM models as well as other baseline implementation.
• utils/ contains utility functions used in data preparation and in different model implementation.
• requirements.txt contains the libraries and tools needed to execute the models.

Dependencies

The models has been tested running under Python 3.9.7, with the required packages installed (see the requirements.txt for package details along with their dependencies). In addition, CUDA 11.1 in a Windows server has been used.

Instructions

Step-1:

Install python and other dependencies; During installation “Disabled path length limit” if needed. Optionally can install jupyter notebook to run on jupyter-lab.

Step-2:

Open terminal/cmd prompt and navigate to the root folder of the downloaded project where “requirements.txt” exists. Now execute this command: pip install -r requirements.txt It will install all the dependencies with the proper version.

Note 1: For some cases, you might be required to execute this command before proceeding to the earlier cmd: pip install wheel If you are using Anaconda or any other kind of python environment where pip cmd is not available then make sure to install the dependencies with the proper version listed in requirements.txt.

Step-3: Once all dependencies installed,navigate to the extracted project's root directory then navigate to the models folder. Then You can run any model file using cmd: python model-file-name.py (such as: python gcn_lstm_model.py)

Note 2: For some cases, you may need to set/modify the dataset path in the model files.

Visualizing embeddings using t-SNE

To investigate the quality of learned MVTS representations, we provide a visualization of t-SNE transformed MVTS representations extracted by the final layer of the GCN-LSTM model.