def _check_graph(x, a, e, y):
g = Graph() # Empty graph
g = Graph(x=x) # Only node features
g = Graph(a=a) # Only adjacency
g = Graph(x=x, a=a, e=e, y=y, extra=1) # Complete graph with extra attribute
# numpy
g_np = g.numpy()
g_gt_names = ["x", "a", "e", "y"]
g_gt = [x, a, e, y]
for i in range(len(g_gt)):
assert np.all(g_np[i] == g_gt[i])
# __getitem__
for i in range(len(g_gt)):
assert np.all(g.__getitem__(g_gt_names[i]) == g_gt[i])
# __repr__
print(g)
# Properties
assert g.n_nodes == n_nodes
assert g.n_node_features == n_node_features
assert g.n_edge_features == n_edge_features
assert g.n_labels == n_out
assert g.n_edges == np.count_nonzero(a)
def read(self):
return [
Graph(x=np.random.rand(n, f),
a=sp.csr_matrix(np.random.randint(0, 2, (n, n))),
e=np.random.rand(n, n, s),
y=np.ones((n, 2))) for n in ns
]
def read(self):
return [
Graph(x=np.random.rand(n, f),
a=sp.csr_matrix(np.random.randint(0, 2, (n, n))),
e=np.random.rand(n, n, s),
y=np.array([0., 1.])) for n in ns
]
def read(self):
return [
Graph(x=np.random.rand(n, f),
a=np.random.randint(0, 2, (n, n)),
e=np.random.rand(n, n, s),
y=np.array([0., 1.])) for n in Ns
]
def read(self):
n = 10
return [
Graph(x=np.random.rand(n, f),
a=sp.csr_matrix(np.random.randint(0, 2, (n, n))),
e=np.random.rand(n, n, s),
y=np.array(n * [[0., 1.]]))
]
def read(self):
n = np.random.randint(3, 8)
self.a = sp.csr_matrix(np.random.randint(0, 2, (n, n)))
return [
Graph(
x=np.random.rand(n, f),
e=np.random.rand(n, n, s),
y=np.array([0.0, 1.0]),
) for _ in range(n_graphs)
]
def test_dataset():
class TestDataset(Dataset):
def read(self):
return [
Graph(x=np.random.rand(n, f),
a=np.random.randint(0, 2, (n, n)),
e=np.random.rand(n, n, s),
y=np.array([0., 1.])) for n in Ns
]
d = TestDataset()
assert d.n_node_features == f
assert d.n_edge_features == s
assert d.n_labels == 2
# signature
for k in ['x', 'a', 'e', 'y']:
assert k in d.signature
# __getitem__
assert isinstance(d[0], Graph)
assert isinstance(d[:3], Dataset)
assert isinstance(d[[1, 3, 4]], Dataset)
# __setitem__
n = 100
g = Graph(x=np.random.rand(n, f),
a=np.random.randint(0, 2, (n, n)),
e=np.random.rand(n, n, s),
y=np.array([0., 1.]))
# single assignment
d[0] = g
assert d[0].n_nodes == n and all([d_.n_nodes != n for d_ in d[1:]])
# Slice assignment
d[1:3] = [g] * 2
assert d[1].n_nodes == n and d[2].n_nodes == n and all(
[d_.n_nodes != n for d_ in d[3:]])
# List assignment
d[[3, 4]] = [g] * 2
assert d[3].n_nodes == n and d[4].n_nodes == n and all(
[d_.n_nodes != n for d_ in d[5:]])
# __len__
assert d.__len__() == n_graphs
# __repr__
print(d)
# Test that shuffling doesn't crash
np.random.shuffle(d)
def _check_graph(x, a, e, y):
g = Graph()
g = Graph(x=x)
g = Graph(a=a)
g = Graph(x=x, a=a, e=e, y=y)
# numpy
g_np = g.numpy()
g_gt_names = ["x", "a", "e", "y"]
g_gt = [x, a, e, y]
for i in range(len(g_gt)):
assert np.all(g_np[i] == g_gt[i])
# __getitem__
for i in range(len(g_gt)):
assert np.all(g.__getitem__(g_gt_names[i]) == g_gt[i])
# __repr__
print(g)
def load_off(filename):
"""
Reads an .off file into a Graph object. Node attributes are the 3d
coordinates of the points, all faces are converted to edges.
:param filename: path to the .off file.
:return: a Graph
"""
lines = open(filename, "r").read().lstrip("OF\n").splitlines()
x, faces = _parse_off(lines)
n = x.shape[0]
row, col = np.vstack((faces[:, :2], faces[:, 1:], faces[:, ::2])).T
adj = sp.csr_matrix((np.ones_like(row), (row, col)), shape=(n, n)).tolil()
adj[col, row] = adj[row, col]
adj = adj.T.tocsr()
adj.data = np.clip(adj.data, 0, 1)
return Graph(x=x, adj=adj)
def test_dataset():
class TestDataset(Dataset):
def read(self):
return [
Graph(
x=np.random.rand(n, f),
a=np.random.randint(0, 2, (n, n)),
e=np.random.rand(n, n, s),
y=np.array([0.0, 1.0]),
) for n in Ns
]
d = TestDataset()
assert d.n_node_features == f
assert d.n_edge_features == s
assert d.n_labels == 2
# signature
for k in ["x", "a", "e", "y"]:
assert k in d.signature
# __getitem__
assert isinstance(d[0], Graph)
assert isinstance(d[:3], Dataset)
assert isinstance(d[[1, 3, 4]], Dataset)
# __setitem__
n = 100
g = Graph(
x=np.random.rand(n, f),
a=np.random.randint(0, 2, (n, n)),
e=np.random.rand(n, n, s),
y=np.array([0.0, 1.0]),
)
# single assignment
d[0] = g
assert d[0].n_nodes == n and all([d_.n_nodes != n for d_ in d[1:]])
# Slice assignment
d[1:3] = [g] * 2
assert (d[1].n_nodes == n and d[2].n_nodes == n
and all([d_.n_nodes != n for d_ in d[3:]]))
# List assignment
d[[3, 4]] = [g] * 2
assert (d[3].n_nodes == n and d[4].n_nodes == n
and all([d_.n_nodes != n for d_ in d[5:]]))
# __len__
assert d.__len__() == n_graphs
# __repr__
print(d)
# Test that shuffling doesn't crash
np.random.shuffle(d)
# Test apply()
t = transforms.NormalizeSphere()
d.apply(t)
# Test filter
d.filter(lambda g: g.n_nodes >= 100)
# Test map
t = lambda g: g.n_nodes
d.map(t, reduce=max)