🆕 dgl.graphbolt

dgl.graphbolt is a dataloading framework for GNN that provides well-defined APIs for each stage of the data pipeline and multiple standard implementations.

APIs

Dataset	An abstract dataset.
Task	An abstract task.
ItemSet	An iterable itemset.
ItemSetDict	An iterable ItemsetDict.
ItemSampler	Item Sampler.
NegativeSampler	A negative sampler used to generate negative samples and return a mix of positive and negative samples.
SubgraphSampler	A subgraph sampler used to sample a subgraph from a given set of nodes from a larger graph.
SampledSubgraph	An abstract class for sampled subgraph.
Feature	Base class for feature.
Feature.read	Read from the feature.
Feature.update	Update the feature.
FeatureStore	Base class for feature store.
FeatureStore.read	Read from the feature store.
FeatureStore.update	Update the feature store.
FeatureFetcher	A feature fetcher used to fetch features for node/edge in graphbolt.
CopyTo	DataPipe that transfers each element yielded from the previous DataPipe to the given device.

DataLoaders

SingleProcessDataLoader	Single process DataLoader.
MultiProcessDataLoader	Multiprocessing DataLoader.

Standard Implementations

OnDiskDataset	An on-disk dataset.
BuiltinDataset	GraphBolt builtin on-disk dataset.
OnDiskMetaData	Metadata specification in YAML.
CSCSamplingGraph	Class for CSC sampling graph.
UniformNegativeSampler	Negative samplers randomly select negative destination nodes for each source node based on a uniform distribution.
NeighborSampler	Neighbor sampler is responsible for sampling a subgraph from given data.
LayerNeighborSampler	Sampler that builds computational dependency of node representations via labor sampling for multilayer GNN from the NeurIPS 2023 paper Layer-Neighbor Sampling – Defusing Neighborhood Explosion in GNNs
SampledSubgraphImpl	Class for sampled subgraph specific for CSCSamplingGraph.
BasicFeatureStore	Basic feature store.
TorchBasedFeature	Torch based feature.
TorchBasedFeatureStore	Torch based feature store.
GPUCachedFeature	GPU cached feature wrapping a fallback feature.