Fit a Graph Attention Network on a WoSIS-style co-location graph

Fit a Graph Attention Network on a WoSIS-style co-location graph

Usage

gnn_fit(
  graph,
  targets,
  hidden = 16L,
  n_heads = 4L,
  n_layers = 2L,
  epochs = 200L,
  lr = 0.01,
  seed = NULL,
  backend = c("r", "torch"),
  device = c("cpu", "mps", "cuda")
)

Arguments

graph

An edaphos_gnn_graph from gnn_build_graph().

targets

Numeric vector, one value per node.

hidden

Integer; output dimension of each GAT layer. Default 16L.

n_heads

Integer; number of attention heads. Default 4L.

n_layers

Integer; number of stacked GAT layers. Default 2L.

epochs

Integer; training epochs for the final linear head.

lr

Numeric; learning rate.

seed

RNG seed.

backend

"r" (default, ELM-style) or "torch" (full autograd with multi-head attention; requires the torch Suggests dependency). v2.7.0 upgrade.

device

"cpu" (default), "mps", or "cuda" when backend = "torch".

Value

An edaphos_gnn_gat fit.