pylibcugraphops.operators.update_efeat_e2e_concat_bwd#

pylibcugraphops.operators.update_efeat_e2e_concat_bwd = <nanobind.nb_func object>#

Computes the backward pass for an operation which concatenates: edge features, and the node features of the source and destination nodes of each each in an edge-to-edge operation.

update_efeat_e2e_concat_bwd(
    grad_input_edge_embedding: Optional[device array],
    grad_input_src_node_embedding: Optional[device array],
    grad_input_dst_node_embedding: Optional[device array],
    grad_output_edge_embedding: device array, graph: pylibcugraphops.bipartite_csc_int[32|64],
    dim_edge: int, dim_src: int, dim_dst: int, stream_id: int = 0
)

Parameters:

grad_input_edge_embeddingdevice array type | None: Device array containing the gradient on the input edge embeddings which is not computed if array is passed as None.
grad_input_src_node_embeddingdevice array type | None: Device array containing the gradient on the input source node embeddings which is not computed if array is passed as None.
grad_input_dst_node_embeddingdevice array type | None: Device array containing the gradient on the input destination node embeddings which is not computed if array is passed as None.
grad_output_edge_embeddingdevice array type: Device array containing the gradient on the output edge embeddings.
graphopaque CSC graph type: graph used for the operation.
dim_edgeint: Dimension of input edge embeddings from the forward pass which can be 0 if the corresponding array has been passed as None during the forward pass.
dim_srcint: Dimension of input source node embeddings from the forward pass which can be 0 if the corresponding array has been passed as None during the forward pass.
dim_dstint: Dimension of input target node embeddings from the forward pass which can be 0 if the corresponding array has been passed as None during the forward pass.
stream_idint, default=0: CUDA stream pointer as a python int