def test_random_walk_edge_chain_pattern(rng, chain_length):
""" Test subsampling a chain sub-graph """
n_iter = max(20, 100 * chain_length)
multiplier = 4
graph = DirectedGraph()
for v in range(chain_length):
for i in range(multiplier):
for j in range(multiplier):
graph.add_edge((v, i), (v + 1, j))
subgraph = random_walk_edge_sample(
graph,
rng,
n_iter,
n_seeds=1, # any starting point in the chain
use_opposite=True,
use_both_ends=True,
max_in_degree=1,
max_out_degree=1)
# Assert graph is a chain of expected length
assert len(subgraph.edges) == chain_length
assert (0 <= d <= 1 for _, d in subgraph.out_degree())
assert (0 <= d <= 1 for _, d in subgraph.in_degree())
# Find vertices - sorting here should retrieve the chaining
vertices = sorted(list(subgraph.nodes))
assert all(v in set((i, m) for m in range(multiplier))
for i, v in enumerate(vertices))
# Inspect linkage
assert all(e in subgraph.edges for e in zip(vertices, vertices[1:]))
def test_random_walk_edge_star_pattern(rng, max_degree):
""" Test subsampling of {1, k} -> 0 and then 0 -> {1, k} """
n_iter = max(100, max_degree * 100)
if max_degree <= 0:
max_degree = 10
graph = DirectedGraph()
for v in range(1, max_degree * 2):
graph.add_edge(v, 0)
subgraph = random_walk_edge_sample(graph,
rng,
n_iter,
n_seeds=1,
use_opposite=False,
use_both_ends=False)
assert len(subgraph.edges) == 1 # only the seed
subgraph = random_walk_edge_sample(graph,
rng,
n_iter,
n_seeds=1,
use_opposite=False,
use_both_ends=True)
assert len(subgraph.edges) == 1 # only the seed
# Here we activate matching of edges *->0
# => we can sample almost all the graph
subgraph = random_walk_edge_sample(graph,
rng,
n_iter,
n_seeds=1,
use_opposite=True,
use_both_ends=False,
max_out_degree=1,
max_in_degree=max_degree)
assert len(subgraph.edges) == min(max_degree, n_iter)
def test_random_walk_edge_functional_pattern(rng):
"""
Minimal test case for the different sampling head and direction
Graph to use should enable precise testing of the options.
4 -> 0 -> 1 -> 2
5 <- 0 1 <- 3
"""
n_iter = 100 # override fixture to stabilize the tests
graph = DirectedGraph()
graph.add_edge(0, 1)
graph.add_edge(1, 2)
graph.add_edge(3, 1)
graph.add_edge(4, 0)
graph.add_edge(0, 5)
# Step 1 - use_opposite=False, use_both_ends=False
# From (0, 1), we should sample only edges starting from 1
expected_subgraph = DirectedGraph()
expected_subgraph.add_edge(0, 1)
expected_subgraph.add_edge(1, 2)
subgraph = random_walk_edge_sample(graph,
rng,
n_iter,
seeds=[(0, 1)],
use_opposite=False,
use_both_ends=False)
assert subgraph == expected_subgraph
# Step 2 - use_opposite=False, use_both_ends=True
# From (0, 1), we should sample only edges starting from 0 or 1
expected_subgraph = DirectedGraph()
expected_subgraph.add_edge(0, 1)
expected_subgraph.add_edge(1, 2)
expected_subgraph.add_edge(0, 5)
subgraph = random_walk_edge_sample(graph,
rng,
n_iter,
seeds=[(0, 1)],
use_opposite=False,
use_both_ends=True)
assert subgraph == expected_subgraph
# Step 3 - use_opposite=True, use_both_ends=False
expected_subgraph = DirectedGraph()
expected_subgraph.add_edge(0, 1)
expected_subgraph.add_edge(1, 2)
expected_subgraph.add_edge(3, 1)
subgraph = random_walk_edge_sample(graph,
rng,
n_iter,
seeds=[(0, 1)],
use_opposite=True,
use_both_ends=False)
assert subgraph == expected_subgraph
# Step 4 - use_opposite=True, use_both_ends=True => catch all
subgraph = random_walk_edge_sample(graph,
rng,
n_iter,
seeds=[(0, 1)],
use_opposite=True,
use_both_ends=True)
assert subgraph == graph