Skip to content

Latest commit

 

History

History
44 lines (26 loc) · 2.62 KB

File metadata and controls

44 lines (26 loc) · 2.62 KB
graph LR
    gearnet_layer_IEConvLayer["gearnet.layer.IEConvLayer"]
    gearnet_layer_GeometricRelationalGraphConv["gearnet.layer.GeometricRelationalGraphConv"]
    gearnet_model_GearNetIEConv["gearnet.model.GearNetIEConv"]
    gearnet_model_FusionNetwork["gearnet.model.FusionNetwork"]
    gearnet_model_GearNetIEConv -- "uses" --> gearnet_layer_IEConvLayer
    gearnet_model_GearNetIEConv -- "uses" --> gearnet_layer_GeometricRelationalGraphConv
    gearnet_model_FusionNetwork -- "composes" --> gearnet_model_GearNetIEConv
Loading

CodeBoardingDemoContact

Details

The gearnet subsystem is designed for molecular representation learning, primarily focusing on graph neural networks. It comprises distinct layers that perform specific graph operations and a model that orchestrates these layers to process molecular structures.

gearnet.layer.IEConvLayer

This component represents an Interaction-Enhanced Convolutional Layer. It's a fundamental building block for processing graph-structured data, specifically designed to incorporate detailed interaction information between nodes (e.g., atoms in a molecule).

Related Classes/Methods: None

gearnet.layer.GeometricRelationalGraphConv

This component implements a Geometric Relational Graph Convolutional layer. It's another core layer type within the GearNet architecture, focusing on incorporating both geometric and relational information during graph convolutions.

Related Classes/Methods: None

gearnet.model.GearNetIEConv

This is the main GearNet model, specifically an Interaction-Enhanced Convolutional Graph Neural Network. It orchestrates multiple GeometricRelationalGraphConv layers and optionally IEConvLayer instances to build a deep graph neural network for molecular representation learning.

Related Classes/Methods: None

gearnet.model.FusionNetwork

This component is a higher-level model designed to combine the outputs of two distinct models: a sequence_model and a structure_model. It's likely used for multi-modal learning, integrating information from different representations (e.g., sequence data and structural data of a molecule).

Related Classes/Methods: None