graph LR
GNN_Model_Implementations["GNN Model Implementations"]
pgl_nn_conv["pgl.nn.conv"]
pgl_nn_pool["pgl.nn.pool"]
pgl_nn_conv_GCNConv["pgl.nn.conv.GCNConv"]
pgl_nn_pool_GraphPool["pgl.nn.pool.GraphPool"]
Message_Passing_Engine["Message Passing Engine"]
GNN_Model_Implementations -- "contains" --> pgl_nn_conv
GNN_Model_Implementations -- "contains" --> pgl_nn_pool
pgl_nn_conv -- "relies on" --> Message_Passing_Engine
pgl_nn_conv -- "is complementary to" --> pgl_nn_pool
pgl_nn_pool -- "is complementary to" --> pgl_nn_conv
pgl_nn_pool -- "contains" --> pgl_nn_pool_GraphPool
pgl_nn_conv_GCNConv -- "relies on" --> Message_Passing_Engine
Message_Passing_Engine -- "supports" --> pgl_nn_conv
Message_Passing_Engine -- "supports" --> pgl_nn_conv_GCNConv
click GNN_Model_Implementations href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/PGL/GNN_Model_Implementations.md" "Details"
click Message_Passing_Engine href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/PGL/Message_Passing_Engine.md" "Details"
The PGL (Paddle Graph Learning) library's neural network (nn) subsystem is designed for building and implementing various Graph Neural Network (GNN) models. At its core, the Message Passing Engine provides the fundamental operations for information propagation across graph structures. This engine underpins the pgl.nn.conv module, which houses diverse graph convolution operations, and the pgl.nn.pool module, responsible for graph pooling techniques. Together, these modules form the GNN Model Implementations component, offering a modular and extensible framework for developing advanced GNN architectures. The design emphasizes clear separation of concerns, allowing for flexible combination of convolution and pooling layers to construct complex graph models.
GNN Model Implementations [Expand]
This top-level component provides a comprehensive set of Graph Neural Network architectures and their fundamental layers. It leverages the Message Passing Engine for core operations and embodies a modular design, allowing for interchangeable convolution, attention, and pooling layers. It is primarily composed of the pgl.nn.conv and pgl.nn.pool modules.
Related Classes/Methods: None
This module aggregates various graph convolution operations. Its responsibility is to define how information from a node's neighbors is aggregated and transformed to update the node's representation, directly utilizing the Message Passing Engine.
Related Classes/Methods:
This module provides implementations for graph pooling operations, responsible for reducing graph size or extracting global graph-level features. It's essential for hierarchical graph representations and graph-level prediction tasks.
Related Classes/Methods:
A concrete implementation of the Graph Convolutional Network layer within pgl.nn.conv. It orchestrates the message passing and aggregation logic specific to GCNs, relying on the Message Passing Engine.
Related Classes/Methods:
A foundational class within pgl.nn.pool for defining generic graph pooling operations. Specific pooling methods would inherit from or utilize this base class.
Related Classes/Methods:
Message Passing Engine [Expand]
This component provides the low-level, efficient primitives for message propagation and aggregation across graph nodes and edges. It forms the backbone upon which all GNN layers are built.
Related Classes/Methods: