Implementation of Graph Locally Linear Embedding (LLE) using Cupy

This repository contains an accurate implementation of "Graph Locally Linear Embedding." The algorithm's details can be found in the paper "Graph Embedding Techniques, Applications, and Performance: A Survey" by Palash Goyal and Emilio Ferrara (2018).

• LLE_Cupy.py: Contains the provided implementation of LLE.

• Demonstration.ipynb: A Jupyter notebook demonstrating our LLE implementation and highlighting the differences from the LLE implementation in the Python package nxt-gem. Ensure the following Python packages are also installed to run this notebook: matplotlib, sklearn, pyspark, networkx, nxt-gem.

• evaluate_embedding.py: Provides code for evaluating graph embeddings. This is also required to run Demonstration.ipynb.

Our implementation of "Locally Linear Embedding" relies on the networkx package. Additionally, the cupy package is required, and a compatible GPU must be available on the machine to run Cupy efficiently. Make sure to check your GPU compatibility with Cupy and install the necessary GPU drivers if needed.

pip install networkx cupy

The lle_cupy(graph, dim) function takes two parameters. The first parameter is a NetworkX graph, and the second parameter is the user-defined embedding dimension. It is important to note that the graph must be connected for the LLE algorithm to be applied successfully.

To utilize the LLE implementation, call the lle_cupy function with a connected NetworkX graph and the desired embedding dimension.

from LLE_Cupy import lle_cupy

# Example usage
graph = ...  # Your connected NetworkX graph
embedding_dimension = ...  # Your desired embedding dimension
embedding_result = lle_cupy(graph, embedding_dimension)