pylibcugraphops.operators.agg_dmpnn_e2e_bwd

pylibcugraphops.operators.agg_dmpnn_e2e_bwd = <nanobind.nb_func object>

Computes the backward pass for D-MPNN-like full graph aggregation

(agg_dmpnn_csc) using edge features in an edge-to-edge reduction scheme (e2e)

agg_dmpnn_e2n_fwd(
    grad_input_edge_embedding: device array,
    grad_output_edge_embedding: device array,
    graph: pylibcugraphops.csc_int[32|64],
    rev_edge_ids: device array,
    concat_own: bool, stream_id: int = 0
)

Parameters:

grad_input_edge_embeddingdevice array type

Device array containing the output gradient on input edge embeddings of forward.

grad_output_edge_embeddingdevice array type

Device array containing the input gradient on output edge embeddings of forward.

graphopaque graph type

graph used for the operation.

rev_edge_idsdevice array

position indices for each edge of the corresponding reverse edge

concat_ownbool

concat an edge’s input features to the aggregated output features

stream_idint, default=0

CUDA stream pointer as a python int