def test_hoptuner_autorgression():
_, G, groups = next(pg.load_datasets_multiple_communities(["bigraph"]))
group = groups[0]
training, evaluation = pg.split(pg.to_signal(G, {v: 1 for v in group}), training_samples=0.01)
auc1 = pg.AUC(evaluation, exclude=training)(pg.HopTuner(measure=pg.AUC).rank(training))
auc3 = pg.AUC(evaluation, exclude=training)(pg.HopTuner(measure=pg.AUC, autoregression=5).rank(training))
assert auc3 > auc1*0.9
def test_hoptuner_explicit_algorithm():
_, G, groups = next(pg.load_datasets_multiple_communities(["bigraph"]))
group = groups[0]
training, evaluation = pg.split(pg.to_signal(G, {v: 1 for v in group}), training_samples=0.5)
auc1 = pg.AUC(evaluation, exclude=training)(pg.HopTuner(lambda params: pg.GenericGraphFilter(params, krylov_dims=10), basis="arnoldi", measure=pg.AUC).rank(training))
auc2 = pg.AUC(evaluation, exclude=training)(pg.HopTuner(basis="arnoldi", krylov_dims=10, measure=pg.AUC).rank(training))
assert abs(auc1-auc2) < 0.005
def test_hoptuner_arnoldi_backends():
_, G, groups = next(pg.load_datasets_multiple_communities(["bigraph"]))
group = groups[0]
training, evaluation = pg.split(pg.to_signal(G, {v: 1 for v in group}), training_samples=0.5)
auc1 = pg.AUC(evaluation, exclude=training)(pg.HopTuner(basis="arnoldi", measure=pg.AUC).rank(training))
auc2 = pg.AUC(evaluation, exclude=training)(pg.HopTuner(basis="arnoldi", measure=pg.AUC, tuning_backend="pytorch").rank(training))
auc3 = pg.AUC(evaluation, exclude=training)(pg.HopTuner(basis="arnoldi", measure=pg.AUC, tuning_backend="tensorflow").rank(training))
assert auc1 == auc2
assert auc1 == auc3
def test_hoptuner_arnoldi():
_, G, groups = next(pg.load_datasets_multiple_communities(["bigraph"]))
group = groups[0]
training, evaluation = pg.split(pg.to_signal(G, {v: 1
for v in group}),
training_samples=0.5)
auc1 = pg.AUC(evaluation,
exclude=training)(pg.HopTuner(measure=pg.AUC).rank(training))
auc2 = pg.AUC(evaluation,
exclude=training)(pg.HopTuner(basis="arnoldi",
measure=pg.AUC).rank(training))
assert abs(auc1 - auc2) < 0.005
def __init__(self,
num_inputs,
num_outputs,
hidden=64,
alpha=0.9,
propagate_on_training=True,
graph_dropout=0):
super().__init__([
tf.keras.layers.Dropout(0.5, input_shape=(num_inputs, )),
tf.keras.layers.Dense(
hidden,
activation=tf.nn.relu,
kernel_regularizer=tf.keras.regularizers.L2(1.E-5)),
tf.keras.layers.Dropout(0.5),
tf.keras.layers.Dense(num_outputs, activation=tf.nn.relu),
])
if isinstance(alpha, str) and alpha == "estimated":
self.ranker = pg.HopTuner(renormalize=True,
assume_immutability=True,
measure=pg.Cos,
tuning_backend="numpy",
tunable_offset=None,
num_parameters=10,
tol=0,
autoregression=5,
error_type="iters",
max_iters=10)
else:
if isinstance(alpha, tf.Variable):
self.trainable_variables = self.trainable_variables + [alpha]
self.ranker = pg.PageRank(alpha,
renormalize=True,
assume_immutability=True,
error_type="iters",
max_iters=10)
self.graph_dropout = graph_dropout
self.num_outputs = num_outputs
self.propagate_on_training = propagate_on_training
def test_autotune_citations():
assert pg.ParameterTuner().cite() != pg.GenericGraphFilter().cite()
assert pg.HopTuner().cite() != pg.GenericGraphFilter().cite()
assert pg.AlgorithmSelection().cite() != pg.GenericGraphFilter().cite()
pg.HeatKernel(t=7,
preprocessor=pre,
max_iters=10000,
tol=tol,
optimization_dict=optimization),
}
algorithms = algorithms # | pg.benchmarks.create_variations(algorithms, {"+sweep": pg.Sweep})
tuned = {
"selected":
pg.AlgorithmSelection(algorithms.values(), fraction_of_training=0.8),
#"tuned": pg.ParameterTuner(preprocessor=pre, fraction_of_training=0.8, tol=tol, optimization_dict=optimization, measure=pg.AUC),
"arnoldi":
pg.HopTuner(preprocessor=pre,
basis="arnoldi",
measure=pg.Cos,
tol=tol,
optimization_dict=optimization),
#"arnoldi2": pg.ParameterTuner(lambda params: pg.HopTuner(preprocessor=pre, basis="arnoldi", num_parameters=int(params[0]),
# measure=pg.Cos,
# tol=tol, optimization_dict=optimization, tunable_offset=None),
# max_vals=[40], min_vals=[5], divide_range=2, fraction_of_training=0.1),
}
#algorithms = pg.create_variations(algorithms, {"": pg.Tautology, "+Sweep": pg.Sweep})
#print(algorithms.keys())
#for name, graph, group in pg.load_datasets_all_communities(datasets, min_group_size=50):
# print(" & ".join([str(val) for val in [name, len(graph), graph.number_of_edges(), len(group)]])+" \\\\")
loader = pg.load_datasets_all_communities(datasets, min_group_size=50)
pg.benchmark_print(pg.benchmark(algorithms | tuned,