def process() -> tuple:
X, Y = dataset_helper.get_dataset_cached(graph_cache_file)
X = graph_helper.get_graphs_only(X)
graph_helper.convert_graphs_to_adjs_tuples(X)
X = [(x, [0] * len(y)) for x, y in X]
estimator, params = task_helper.get_graph_estimator_and_params(X, Y)
return ClassificationData(X, Y, estimator, params)
def transform(self, X, y=None, **fit_params):
if self.from_nx_to_tuple:
X = graph_helper.get_graphs_only(X)
graph_helper.convert_graphs_to_adjs_tuples(X)
else:
graph_helper.convert_adjs_tuples_to_graphs(X)
return X
def test_convert_graph_datasets(self):
for graph_dataset, dataset_name in self.iterate_graph_cache_datasets():
X, Y = dataset_helper.get_dataset_cached(graph_dataset)
self.assertTrue(len(X))
self.assertTrue(len(Y))
graph_helper.convert_graphs_to_adjs_tuples(X)
for x in X:
self.assertTrue(isinstance(x, tuple))
self.assertTrue(isinstance(x[0], scipy.sparse.spmatrix))
self.assertTrue(isinstance(x[1], list))
break
def get_graphs(as_adj=False):
g_train = nx.Graph()
g_train.add_edge('A', 'B')
g_test = nx.Graph()
g_test.add_edge('A', 'C')
if as_adj:
g_test = graph_helper.convert_graphs_to_adjs_tuples([g_test],
copy=True)[0]
g_train = graph_helper.convert_graphs_to_adjs_tuples([g_train],
copy=True)[0]
return g_train, g_test
def get_graph_estimator_and_params(X,
Y=None,
reduced: bool = False,
with_node_weights: bool = False,
add_phi_dim=False):
assert len(X)
X_ = X
estimator, params = graph_pipeline.get_params(
reduced=reduced, with_node_weights=with_node_weights)
if add_phi_dim:
if isinstance(X[0], nx.Graph):
X_ = graph_helper.convert_graphs_to_adjs_tuples(X, copy=True)
assert isinstance(X_[0], tuple) and isinstance(X_[0][1], list)
num_vertices = get_num_vertices(X_) * 5
graph_pipeline.add_num_vertices_to_fast_wl_params(params, num_vertices)
return estimator, params
def _retrieve_node_weights_and_convert_graphs(X,
node_weight_function=None,
same_label=False,
use_directed=True,
use_nx=True):
X = graph_helper.get_graphs_only(X)
if not use_directed:
if use_nx:
X = [nx.Graph(x) for x in X]
assert not np.any([x.is_directed() for x in X])
else:
raise NotImplementedError(
'!use_directed and !use_nx not implemented')
node_weight_factors = get_node_weight_factors(X,
metric=node_weight_function)
X = graph_helper.convert_graphs_to_adjs_tuples(X, copy=True)
if same_label:
X = [(adj, ['dummy'] * len(labels)) for adj, labels in X]
return X, node_weight_factors
from time import time
from utils import dataset_helper, graph_helper, primes, time_utils
DATASET = 'ng20'
H = 10
USED_SPARSE_MATRIX = scipy.sparse.lil_matrix
PERF_LOG = './perf.log'
# Warm up primes cache
primes.get_highest_prime_range()
# Get dataset
X, Y = dataset_helper.get_concept_map_for_dataset(DATASET, graphs_only=True)
# Convert to adj
X = graph_helper.convert_graphs_to_adjs_tuples(X, copy=True)
# Same label for all nodes
adj, labels = zip(*X)
labels = [[1] * len(x) for x in labels]
X = list(zip(adj, labels))
# Start profiling
thread = flamegraph.start_profile_thread(fd=open(PERF_LOG, "w"))
start = time()
# Calculate fast_wl
phi_lists, new_label_lookups, new_label_counters = fast_wl.transform(
X, h=H, used_matrix_type=USED_SPARSE_MATRIX)
# Convert to sparse lil_matrix for "fancy" indexing (needed for other pipeline elements)