Download tests/test_graphs.py from HWresearch/GNN4Colliders: direct link, hf CLI and curl.
- Browser
- Download file 2.38 kB
-
https://huggingface.co/HWresearch/GNN4Colliders/resolve/main/tests/test_graphs.py
- Command line
-
hf download hf://HWresearch/GNN4Colliders/tests/test_graphs.py
-
curl -L -o test_graphs.py https://huggingface.co/HWresearch/GNN4Colliders/resolve/main/tests/test_graphs.py
2.38 kB
| import math | |
| import pytest | |
| import torch | |
| from conftest import random_events | |
| from gnn4colliders.data import pack_events | |
| from gnn4colliders.graphs import ( | |
| PADDING_GRAPHS, | |
| StaticGraphBatch, | |
| collate_graph_batch, | |
| edge_features, | |
| fully_connected_edges, | |
| graph_batch_capacity, | |
| ) | |
| from gnn4colliders.models import EdgeNetwork | |
| def test_fully_connected_edges_are_source_major_without_self_loops(): | |
| src, dst, counts = fully_connected_edges(torch.tensor([3, 1, 2])) | |
| assert counts.tolist() == [6, 1, 2] | |
| pairs = list(zip(src.tolist(), dst.tolist())) | |
| assert pairs[:6] == [(0, 1), (0, 2), (1, 0), (1, 2), (2, 0), (2, 1)] | |
| # A one-node graph keeps its single self-loop. | |
| assert pairs[6:] == [(3, 3), (4, 5), (5, 4)] | |
| def test_edge_features_wrap_delta_phi(): | |
| nodes = torch.zeros((2, 7)) | |
| nodes[:, 1] = torch.tensor([0.5, -0.5]) | |
| nodes[:, 2] = torch.tensor([3.0, -3.0]) | |
| features = edge_features(nodes, torch.tensor([0]), torch.tensor([1])) | |
| deta, dphi, dr = features[0].tolist() | |
| assert deta == pytest.approx(1.0) | |
| assert dphi == pytest.approx(6.0 - 2 * math.pi) | |
| assert dr == pytest.approx(math.hypot(deta, dphi)) | |
| def test_padded_batch_reproduces_real_graph_outputs(): | |
| events = random_events(12) | |
| batch = collate_graph_batch(events, torch.arange(len(events))) | |
| torch.manual_seed(0) | |
| model = EdgeNetwork(7, 3, 0, 16, 3, 2, 2).eval() | |
| static = StaticGraphBatch( | |
| batch, graph_batch_capacity(batch, node_multiple=32, edge_multiple=64) | |
| ) | |
| # Garbage left in padding rows by an earlier, larger batch must not leak. | |
| static.graph.node_features.normal_() | |
| static.graph.edge_features.normal_() | |
| static.load(batch) | |
| with torch.no_grad(): | |
| expected, padded = model(batch), model(static.batch) | |
| assert padded.shape[0] == len(events) + PADDING_GRAPHS | |
| torch.testing.assert_close(padded[: len(events)], expected, rtol=1e-5, atol=1e-6) | |
| def test_static_batch_rejects_batches_beyond_capacity(): | |
| def batch(nodes: int): | |
| events = pack_events([torch.rand((nodes, 7)) for _ in range(4)], [0] * 4) | |
| return collate_graph_batch(events, torch.arange(4)) | |
| small = batch(1) | |
| static = StaticGraphBatch( | |
| small, graph_batch_capacity(small, node_multiple=1, edge_multiple=1) | |
| ) | |
| with pytest.raises(ValueError, match="capacity"): | |
| static.load(batch(5)) | |