Fit disjoint graph regression

// Three disjoint graphs, two training targets and one held-out graph target.
import aner.nn

let graphs = nn.graph(nodes: 6,
    sources: List(0, 1, 2, 3, 4, 5),
    destinations: List(1, 0, 3, 2, 5, 4),
    graph_ids: List(0, 0, 1, 1, 2, 2))
let features: Tensor = [[-1.0], [-0.6], [0.6], [1.0], [0.2], [0.4]]
let targets: Tensor = [[-0.8], [0.8], [0.3]]
var model = nn.gnn(1, 8, 1, architecture: "gin", layers: 1)
    .task("graph_regression").on("cpu", "Float64")
var step = 0
while step < 30 {
    model = model.train_step(graphs, features, targets,
        rate: 0.01, indices: List(0, 1))
    step = step + 1
}
let validation = model.evaluate(graphs, features, targets, indices: List(2))
print(validation.summary())
print(validation.samples())
print(graphs.graph_count())
print(model.updates())