API REFERENCE/omnibias-graph

omnibias.graph.arrangement.torch

0.1.0a2AlphaPython >=3.10AGPL / commercial

Spectral graph operators and differentiable relaxations for ordering, assignment, and graph-structured learning.

import omnibias.graph.arrangement.torch

Module contract & mathematical conventions

Face-Net (torch; theory 02-02). Temperature collapse; subgraph sampling.

Shared formulas do not imply universal floating-point identity. Verify dtype, shape, derivative-order support, and the stated numerical assumptions for the path you use.

Module source ↗Source revision 9a2d73df0573Running reference examples →

CLASS

FaceNet

omnibias.graph.arrangement.FaceNetSource ↗

class FaceNet(nn.Module)

Message passing on a discovered arrangement graph. Not a complete lattice.

Parameters

Name	Type & default	Contract
arrangement_dimpositional-or-keyword	intRequired	See the operation contract and implementation below.
hiddenpositional-or-keyword	intRequired	See the operation contract and implementation below.
roundspositional-or-keyword	intRequired	See the operation contract and implementation below.
betakeyword-only	floatDefault: 5.0	See the operation contract and implementation below.
dtypekeyword-only	torch.dtype | NoneDefault: None	See the operation contract and implementation below.

Executed regression · g6 torch runs

Recorded coverage confirms this test executed the API body. This example belongs to the pinned upstream test suite. It may use fixtures and imports from its full file. Run it from that checkout with its test dependencies installed.

Full file, imports & fixtures ↗

def test_g6_torch_runs() -> None:
    pytest.importorskip("torch")
    import torch
    from omnibias.graph.arrangement.torch import FaceNet

    torch.set_default_dtype(torch.float64)
    arr = _square()
    samples = np.array([[0.0, 0.0], [2.0, 0.0], [0.0, 2.0], [-2.0, 0.0], [0.0, -2.0]])
    graph = build_arrangement_graph(arr, samples)
    net = FaceNet(2, hidden=4, rounds=2, dtype=torch.float64)
    out = net(graph, arr)
    assert out.shape[0] == len(graph.cells)
    assert torch.isfinite(out).all()

python -m pytest "packages/omnibias-graph/tests/test_arrangement.py::test_g6_torch_runs" -q

Fields & methods

__init__method

__init__(self, arrangement_dim: int, hidden: int, rounds: int, *, beta: float=5.0, dtype: torch.dtype | None=None) -> None

See the class contract above and the exact method implementation below.

Parameters

Name	Type & default	Contract
arrangement_dimpositional-or-keyword	intRequired	See the operation contract and implementation below.
hiddenpositional-or-keyword	intRequired	See the operation contract and implementation below.
roundspositional-or-keyword	intRequired	See the operation contract and implementation below.
betakeyword-only	floatDefault: 5.0	See the operation contract and implementation below.
dtypekeyword-only	torch.dtype | NoneDefault: None	See the operation contract and implementation below.

Returns: None

Validation & failure cases

• When rounds < 1, raises ValueError('rounds must be >= 1').

Executed regression · g6 torch runs

Recorded coverage confirms this test executed the API body. This example belongs to the pinned upstream test suite. It may use fixtures and imports from its full file. Run it from that checkout with its test dependencies installed.

Full file, imports & fixtures ↗

def test_g6_torch_runs() -> None:
    pytest.importorskip("torch")
    import torch
    from omnibias.graph.arrangement.torch import FaceNet

    torch.set_default_dtype(torch.float64)
    arr = _square()
    samples = np.array([[0.0, 0.0], [2.0, 0.0], [0.0, 2.0], [-2.0, 0.0], [0.0, -2.0]])
    graph = build_arrangement_graph(arr, samples)
    net = FaceNet(2, hidden=4, rounds=2, dtype=torch.float64)
    out = net(graph, arr)
    assert out.shape[0] == len(graph.cells)
    assert torch.isfinite(out).all()

python -m pytest "packages/omnibias-graph/tests/test_arrangement.py::test_g6_torch_runs" -q

Method implementation

    def __init__(
        self,
        arrangement_dim: int,
        hidden: int,
        rounds: int,
        *,
        beta: float = 5.0,
        dtype: torch.dtype | None = None,
    ) -> None:
        super().__init__()
        if rounds < 1:
            raise ValueError("rounds must be >= 1")
        self.rounds = int(rounds)
        self.beta = float(beta)
        dt = torch.get_default_dtype() if dtype is None else dtype
        in_dim = 2 + int(arrangement_dim)
        self.lift = nn.Linear(in_dim, hidden, dtype=dt)
        self.msg = nn.Linear(hidden + 1, hidden, dtype=dt)
        self.readout = nn.Linear(hidden, 1, dtype=dt)

View source ↗

forwardmethod

forward(self, graph: ArrangementGraph, arr: Arrangement) -> Tensor

See the class contract above and the exact method implementation below.

Parameters

Name	Type & default	Contract
graphpositional-or-keyword	ArrangementGraphRequired	See the operation contract and implementation below.
arrpositional-or-keyword	ArrangementRequired	See the operation contract and implementation below.

Returns: Tensor

Executed regression · g6 torch runs

Recorded coverage confirms this test executed the API body. This example belongs to the pinned upstream test suite. It may use fixtures and imports from its full file. Run it from that checkout with its test dependencies installed.

Full file, imports & fixtures ↗

def test_g6_torch_runs() -> None:
    pytest.importorskip("torch")
    import torch
    from omnibias.graph.arrangement.torch import FaceNet

    torch.set_default_dtype(torch.float64)
    arr = _square()
    samples = np.array([[0.0, 0.0], [2.0, 0.0], [0.0, 2.0], [-2.0, 0.0], [0.0, -2.0]])
    graph = build_arrangement_graph(arr, samples)
    net = FaceNet(2, hidden=4, rounds=2, dtype=torch.float64)
    out = net(graph, arr)
    assert out.shape[0] == len(graph.cells)
    assert torch.isfinite(out).all()

python -m pytest "packages/omnibias-graph/tests/test_arrangement.py::test_g6_torch_runs" -q

Method implementation

    def forward(self, graph: ArrangementGraph, arr: Arrangement) -> Tensor:
        feats = torch.tensor(node_features(arr, graph, beta=self.beta), dtype=self.lift.weight.dtype)
        h = torch.tanh(self.lift(feats))
        n = h.shape[0]
        adj = torch.zeros(n, n, dtype=h.dtype)
        crossed = torch.zeros(n, n, dtype=h.dtype)
        for u, v, k in graph.edges:
            adj[u, v] = adj[v, u] = 1.0
            crossed[u, v] = crossed[v, u] = float(k)
        for _ in range(self.rounds):
            msgs = []
            for i in range(n):
                acc = torch.zeros_like(h[i])
                deg = 0.0
                for j in range(n):
                    if float(adj[i, j].detach()) <= 0.0:
                        continue
                    cat = torch.cat([h[j], crossed[i, j].reshape(1)])
                    acc = acc + torch.tanh(self.msg(cat))
                    deg += 1.0
                msgs.append(acc / max(deg, 1.0))
            h = h + torch.stack(msgs)
        return self.readout(h).squeeze(-1)

View source ↗

roundsattribute

rounds = int(rounds)

Public instance attribute assigned by __init__. The source assignment is rounds = int(rounds); it describes instance state, not a literal constructor default.

View source ↗

betaattribute

beta = float(beta)

Public instance attribute assigned by __init__. The source assignment is beta = float(beta); it describes instance state, not a literal constructor default.

View source ↗

liftattribute

lift = nn.Linear(in_dim, hidden, dtype=dt)

Public instance attribute assigned by __init__. The source assignment is lift = nn.Linear(in_dim, hidden, dtype=dt); it describes instance state, not a literal constructor default.

View source ↗

msgattribute

msg = nn.Linear(hidden + 1, hidden, dtype=dt)

Public instance attribute assigned by __init__. The source assignment is msg = nn.Linear(hidden + 1, hidden, dtype=dt); it describes instance state, not a literal constructor default.

View source ↗

readoutattribute

readout = nn.Linear(hidden, 1, dtype=dt)

Public instance attribute assigned by __init__. The source assignment is readout = nn.Linear(hidden, 1, dtype=dt); it describes instance state, not a literal constructor default.

View source ↗

Read the implementation

class FaceNet(nn.Module):
    """Message passing on a discovered arrangement graph. Not a complete lattice."""

    def __init__(
        self,
        arrangement_dim: int,
        hidden: int,
        rounds: int,
        *,
        beta: float = 5.0,
        dtype: torch.dtype | None = None,
    ) -> None:
        super().__init__()
        if rounds < 1:
            raise ValueError("rounds must be >= 1")
        self.rounds = int(rounds)
        self.beta = float(beta)
        dt = torch.get_default_dtype() if dtype is None else dtype
        in_dim = 2 + int(arrangement_dim)
        self.lift = nn.Linear(in_dim, hidden, dtype=dt)
        self.msg = nn.Linear(hidden + 1, hidden, dtype=dt)
        self.readout = nn.Linear(hidden, 1, dtype=dt)

    def forward(self, graph: ArrangementGraph, arr: Arrangement) -> Tensor:
        feats = torch.tensor(node_features(arr, graph, beta=self.beta), dtype=self.lift.weight.dtype)
        h = torch.tanh(self.lift(feats))
        n = h.shape[0]
        adj = torch.zeros(n, n, dtype=h.dtype)
        crossed = torch.zeros(n, n, dtype=h.dtype)
        for u, v, k in graph.edges:
            adj[u, v] = adj[v, u] = 1.0
            crossed[u, v] = crossed[v, u] = float(k)
        for _ in range(self.rounds):
            msgs = []
            for i in range(n):
                acc = torch.zeros_like(h[i])
                deg = 0.0
                for j in range(n):
                    if float(adj[i, j].detach()) <= 0.0:
                        continue
                    cat = torch.cat([h[j], crossed[i, j].reshape(1)])
                    acc = acc + torch.tanh(self.msg(cat))
                    deg += 1.0
                msgs.append(acc / max(deg, 1.0))
            h = h + torch.stack(msgs)
        return self.readout(h).squeeze(-1)

Related APIsomnibias.graph.arrangement.node_features↗

Other public import paths

• omnibias.graph.arrangement.FaceNet
• omnibias.graph.arrangement.torch.FaceNet

← Package overviewRunnable learning examples →