OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Egnn Pytorch

Egnn Pytorch is a Python and PyTorch implementation of E(n)-Equivariant Graph Neural Networks, with layer and network interfaces for processing graph node features, coordinates, and optional edge data.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Egnn Pytorch — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when a Python/PyTorch experiment needs an EGNN implementation that updates both graph features and coordinates, particularly when the model must incorporate edge data or explicit sparse-neighbor structure.

Who it is for

  • Researchers exploring equivariant graph neural network architectures
  • Model developers working with geometric graph data
  • Practitioners prototyping molecular activity prediction or dynamical system models

Documented uses

  • Apply EGNN layers to node features and coordinates
  • Build a deeper network with configurable dimensions, depth, and neighbor selection
  • Pass continuous edge features into graph-learning layers
  • Restrict message passing to an adjacency matrix
  • Explore coordinate normalization and coordinate-weight clamping for configurations with many neighbors
Before you choose

Requirements & limits

The public repository identifies the project as MIT-licensed and provides a pip installation command.

  • The documented interface expects PyTorch tensors for features and coordinates, with optional edge tensors and masks.
  • The network supports nearest-neighbor configuration and adjacency-based sparse message passing.
  • The README flags a masking-related neighbor-selection bug in versions earlier than 0.1.12.
  • For high neighbor counts, the README discusses coordinate normalization and clamping coordinate weights as stability measures.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up and pass graph tensors through an EGNN layer

Install the package and use the documented layer interface to update node features and coordinates.

Before you start

  • Python environment with PyTorch available
  • Node feature and coordinate tensors; optional edge tensor when using edge features

Steps

  1. Install the package with the command below.

    Documented command
    pip install egnn-pytorch

  2. Import EGNN from egnn_pytorch and create a layer with the feature dimension.

  3. Pass feature and coordinate tensors to the layer; if using edges, set edge_dim and pass the edge tensor.

  4. Use the returned feature and coordinate tensors as the updated representations.

Expected outcome

A documented Python workflow for applying an EGNN layer to graph feature and coordinate tensors.

Guide limitations

  • The example demonstrates tensor passing, not model training or evaluation.
  • For masked experiments using versions earlier than 0.1.12, the README warns of a neighbor-selection bug and recommends rerunning those experiments.
Documentation 1 ↗
Source-based practical guide

Configure sparse-neighbor network inputs

Set up the documented network interface to pass an adjacency matrix for sparse message passing.

Before you start

  • PyTorch environment and EGNN_Network from egnn_pytorch
  • Token features, coordinates, a boolean mask, and an adjacency matrix

Steps

  1. Create an EGNN_Network with only_sparse_neighbors=True.

  2. Prepare token features, coordinates, and a boolean mask with matching graph positions.

  3. Prepare an adjacency matrix describing the allowed neighbors.

  4. Pass the features, coordinates, mask, and adjacency matrix to the network.

Expected outcome

A network call configured to attend only to neighbors represented by the supplied adjacency matrix.

Guide limitations

  • The README's example uses a simple chain adjacency; it does not provide a domain-specific graph-construction procedure.
  • This guide covers the documented input setup and call, not training or deployment.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
14 October 2026
Original website title
lucidrains/egnn-pytorch · Official homepage title · retained in its source language
How we review sourcesSuggest a correction