5. Supported operations

This chapter contains a list of the PyTorch Geometric operations supported by the Poplar SDK.

Table 5.1 Supported operations

Layers kind	Layer name	Status	Notes
Basic blocks	Linear	Supported
	Sequential	Supported
	HeteroLinear	Supported
Convolution Layers	SimpleConv	Supported
	ChebConv	Not supported
	SAGEConv	Supported
	CuGraphSAGEConv	Not supported
	GraphConv	Supported
	GatedGraphConv	Supported
	ResGatedGraphConv	Supported
	CuGraphGATConv	Not supported
	FusedGATConv	Not supported
	TransformerConv	Supported
	TAGConv	Supported
	GINConv	Supported
	GINEConv	Supported
	ARMAConv	Supported
	APPNP	Supported
	MFConv	Supported
	RGCNConv	Not supported
	CuGraphRGCNConv	Not supported
	SplineConv	Supported	Only inference is supported.
	NNConv	Supported
	CGConv	Supported
	EdgeConv	Supported
	DynamicEdgeConv	Supported
	XConv	Supported
	HypergraphConv	Supported	num_edges is required.
	LEConv	Supported
	PNAConv	Supported
	ClusterGCNConv	Supported
	PANConv	Not supported
	WLConv	Not supported
	WLConvContinuous	Supported
	SuperGATConv	Not supported
	GeneralConv	Supported
	HGTConv	Supported
	HeteroConv	Supported
	LGConv	Supported
	PointGNNConv	Supported
	GPSConv	Not supported
	RGATConv	Supported	The attention_mechanism option within-relation is not supported.
	FiLMConv	Supported	num_relations cannot be greater than 1.
	GCNConv	Supported	add_self_loops must be set to False.
	GravNetConv
	GATConv
	GATv2Conv
	AGNNConv
	SGConv
	SSGConv
	FastRGCNConv
	SignedConv
	DNAConv
	PointNetConv
	GMMConv
	PPFConv
	FeaStConv
	PointTransformerConv
	GENConv
	GCN2Conv
	FAConv
	EGConv
	PDNConv
	HEATConv
	HANConv
	AntiSymmetricConv
Aggregation Layers	MultiAggregation	Supported	dim_size is required.
	SumAggregation
	MeanAggregation
	MaxAggregation
	MinAggregation
	MulAggregation
	VarAggregation
	StdAggregation
	SoftmaxAggregation
	PowerMeanAggregation
	DeepSetsAggregation
	AttentionalAggregation
	Set2Set
	DegreeScalerAggregation
	MedianAggregation
	QuantileAggregation
	SortAggregation	Supported	dim_size is required. max_num_elements is required.
	LSTMAggregation
	MLPAggregation
	GRUAggregation
	SetTransformerAggregation	Not supported
	GraphMultisetTransformer	Not supported
	EquilibriumAggregation	Not supported
Normalization layers	BatchNorm	Supported
	HeteroBatchNorm	Not supported
	HeteroLayerNorm	Not supported
	InstanceNorm	Supported	batch_size is required.
	GraphNorm
	LayerNorm
	GraphSizeNorm
	PairNorm
	MeanSubtractionNorm	Supported	dim_size is required.
	MessageNorm	Supported
	DiffGroupNorm	Supported
Pooling layers	Pooling	Supported
	global_add_pool	Supported	size is required.
	global_mean_pool
	global_max_pool
	TopKPooling	Not supported
	PANPooling	Not supported
	SAGPooling	Not supported
	EdgePooling	Not supported
	ASAPooling	Not supported
	MemPooling	Supported	batch_size is required. max_num_nodes is required.
	max_pool	Not supported
	avg_pool	Not supported
	consecutive_cluster	Not supported
	max_pool_neighbor_x	Not supported
	avg_pool_neighbor_x	Not supported
	max_pool_x	Supported	batch_size is required. size is required.
	avg_pool_x
	graclus	Not supported
	voxel_grid	Supported
	fps	Supported	ptr is required. batch has to be None.
	knn	Supported
	knn_graph	Supported	loop must be set to True.
	radius	Supported	batch_size is required.
	radius_graph
	nearest	Supported	torch_cluster.nearest has to be replaced with poptorch.nearest. poptorch.nearest supports arguments batch_x and batch_y in the original form of torch.Tensor plus a regular list. Validation of batch indices is NOT performed if batch_x and batch_y are passed as torch.Tensor.
	decimation_indices	Not supported
Unspooling layers	knn_interpolate	Supported
Functional	bro	Not supported
	gini	Supported
Dense convolutional layers	DenseGCNConv	Supported
	DenseGINConv	Supported
	DenseGraphConv	Supported
	DenseSAGEConv	Supported
	DenseGATConv	TBD
Dense pooling layers	dense_diff_pool	Not supported
	dense_mincut_pool	Not supported
	DMoNPooling	Not supported
Encoding	PositionalEncoding	Supported
	TemporalEncoding	Supported
KGE Models	TransE	Supported
	ComplEx	Supported
	DistMult	Supported
	RotatE	Supported