def test_algorithm_selection():
for _ in supported_backends():
_, graph, communities = next(
pg.load_datasets_multiple_communities(["bigraph"],
max_group_number=3))
train, test = pg.split(communities,
0.05) # 5% of community members are known
algorithms = pg.create_variations(pg.create_demo_filters(),
pg.Normalize)
supervised_algorithm = pg.AlgorithmSelection(algorithms.values(),
measure=pg.AUC,
tuning_backend="numpy")
print(supervised_algorithm.cite())
modularity_algorithm = pg.AlgorithmSelection(
algorithms.values(),
fraction_of_training=1,
measure=pg.Modularity().as_supervised_method(),
tuning_backend="numpy")
supervised_aucs = list()
modularity_aucs = list()
for seeds, members in zip(train.values(), test.values()):
measure = pg.AUC(members, exclude=seeds)
supervised_aucs.append(measure(supervised_algorithm(graph, seeds)))
modularity_aucs.append(measure(modularity_algorithm(graph, seeds)))
assert abs(
sum(supervised_aucs) / len(supervised_aucs) -
sum(modularity_aucs) / len(modularity_aucs)) < 0.05
def test_explicit_citations():
assert "unknown node ranking algorithm" == pg.NodeRanking().cite()
assert "with parameters tuned \cite{krasanakis2021pygrank}" in pg.ParameterTuner(
lambda params: pg.PageRank(params[0])).cite()
assert "Postprocessor" in pg.Postprocessor().cite()
assert pg.PageRank().cite() in pg.AlgorithmSelection().cite()
assert "krasanakis2021pygrank" in pg.ParameterTuner().cite()
assert "ortega2018graph" in pg.ParameterTuner().cite()
assert pg.HeatKernel().cite() in pg.SeedOversampling(pg.HeatKernel()).cite()
assert pg.AbsorbingWalks().cite() in pg.BoostedSeedOversampling(pg.AbsorbingWalks()).cite()
assert "krasanakis2018venuerank" in pg.BiasedKernel(converge_to_eigenvectors=True).cite()
assert "yu2021chebyshev" in pg.HeatKernel(coefficient_type="chebyshev").cite()
assert "susnjara2015accelerated" in pg.HeatKernel(krylov_dims=5).cite()
assert "krasanakis2021pygrank" in pg.GenericGraphFilter(optimization_dict=dict()).cite()
assert "tautology" in pg.Tautology().cite()
assert pg.PageRank().cite() == pg.Tautology(pg.PageRank()).cite()
assert "mabs" in pg.MabsMaintain(pg.PageRank()).cite()
assert "max normalization" in pg.Normalize(pg.PageRank()).cite()
assert "[0,1] range" in pg.Normalize(pg.PageRank(), "range").cite()
assert "ordinal" in pg.Ordinals(pg.PageRank()).cite()
assert "exp" in pg.Transformer(pg.PageRank()).cite()
assert "0.5" in pg.Threshold(pg.PageRank(), 0.5).cite()
assert "andersen2007local" in pg.Sweep(pg.PageRank()).cite()
assert pg.HeatKernel().cite() in pg.Sweep(pg.PageRank(), pg.HeatKernel()).cite()
assert "LFPRO" in pg.AdHocFairness("O").cite()
assert "LFPRO" in pg.AdHocFairness(pg.PageRank(), "LFPRO").cite()
assert "multiplicative" in pg.AdHocFairness(pg.PageRank(), "B").cite()
assert "multiplicative" in pg.AdHocFairness(pg.PageRank(), "mult").cite()
assert "tsioutsiouliklis2020fairness" in pg.AdHocFairness().cite()
assert "rahman2019fairwalk" in pg.FairWalk(pg.PageRank()).cite()
assert "krasanakis2020prioredit" in pg.FairPersonalizer(pg.PageRank()).cite()
def test_all_communities_benchmarks():
datasets = ["bigraph"]
pre = pg.preprocessor(assume_immutability=True, normalization="symmetric")
tol = 1.E-9
optimization = pg.SelfClearDict()
algorithms = {
"ppr0.85": pg.PageRank(alpha=0.85, preprocessor=pre, max_iters=10000, tol=tol),
"ppr0.9": pg.PageRank(alpha=0.9, preprocessor=pre, max_iters=10000, tol=tol),
"ppr0.99": pg.PageRank(alpha=0.99, preprocessor=pre, max_iters=10000, tol=tol),
"hk3": pg.HeatKernel(t=3, preprocessor=pre, max_iters=10000, tol=tol, optimization_dict=optimization),
"hk5": pg.HeatKernel(t=5, preprocessor=pre, max_iters=10000, tol=tol, optimization_dict=optimization),
"hk7": pg.HeatKernel(t=7, preprocessor=pre, max_iters=10000, tol=tol, optimization_dict=optimization),
}
tuned = {"selected": pg.AlgorithmSelection(algorithms.values(), fraction_of_training=0.8)}
loader = pg.load_datasets_all_communities(datasets, min_group_size=50)
pg.benchmark_print(pg.benchmark(algorithms | tuned, loader, pg.AUC, fraction_of_training=.8, seed=list(range(1))),
decimals=3, delimiter=" & ", end_line="\\\\")
loader = pg.load_datasets_all_communities(datasets, min_group_size=50)
pg.benchmark_print(pg.benchmark(algorithms | tuned, loader, pg.Modularity, sensitive=pg.pRule, fraction_of_training=.8, seed=list(range(1))),
decimals=3, delimiter=" & ", end_line="\\\\")
mistreatment = lambda known_scores, sensitive_signal, exclude: \
pg.AM([pg.Disparity([pg.TPR(known_scores, exclude=1 - (1 - exclude.np) * sensitive_signal.np),
pg.TPR(known_scores, exclude=1 - (1 - exclude.np) * (1 - sensitive_signal.np))]),
pg.Disparity([pg.TNR(known_scores, exclude=1 - (1 - exclude.np) * sensitive_signal.np),
pg.TNR(known_scores, exclude=1 - (1 - exclude.np) * (1 - sensitive_signal.np))])])
loader = pg.load_datasets_all_communities(datasets, min_group_size=50)
pg.benchmark_print(pg.benchmark(algorithms | tuned, loader, pg.Modularity, sensitive=mistreatment, fraction_of_training=.8, seed=list(range(1))),
decimals=3, delimiter=" & ", end_line="\\\\")
def test_autotune_methods():
import numpy as np
_, 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}))
aucs = [pg.AUC(evaluation, exclude=training)(ranker.rank(training)) for ranker in pg.create_demo_filters().values()]
auc2 = pg.AUC(evaluation, exclude=training)(pg.AlgorithmSelection().rank(training))
assert max(aucs)-np.std(aucs) <= auc2
def test_multigroup_benchmarks():
datasets = ["bigraph"]
pre = pg.preprocessor(assume_immutability=True, normalization="symmetric")
tol = 1.E-9
optimization = pg.SelfClearDict()
algorithms = {
"ppr0.85":
pg.PageRank(alpha=0.85, preprocessor=pre, max_iters=10000, tol=tol),
"ppr0.9":
pg.PageRank(alpha=0.9, preprocessor=pre, max_iters=10000, tol=tol),
"ppr0.99":
pg.PageRank(alpha=0.99, preprocessor=pre, max_iters=10000, tol=tol),
"hk3":
pg.HeatKernel(t=3,
preprocessor=pre,
max_iters=10000,
tol=tol,
optimization_dict=optimization),
"hk5":
pg.HeatKernel(t=5,
preprocessor=pre,
max_iters=10000,
tol=tol,
optimization_dict=optimization),
"hk7":
pg.HeatKernel(t=7,
preprocessor=pre,
max_iters=10000,
tol=tol,
optimization_dict=optimization),
}
tuned = {
"selected":
pg.AlgorithmSelection(algorithms.values(), fraction_of_training=0.8)
}
loader = pg.load_datasets_multiple_communities(datasets, min_group_size=50)
pg.benchmark_print(pg.benchmark(
algorithms | tuned,
loader,
lambda ground_truth, exclude: pg.MultiSupervised(
pg.AUC, ground_truth, exclude),
fraction_of_training=.8,
seed=list(range(1))),
decimals=3,
delimiter=" & ",
end_line="\\\\")
loader = pg.load_datasets_multiple_communities(datasets, min_group_size=50)
pg.benchmark_print(pg.benchmark(algorithms | tuned,
loader,
pg.Modularity,
sensitive=pg.pRule,
fraction_of_training=.8,
seed=list(range(1))),
decimals=3,
delimiter=" & ",
end_line="\\\\")
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()
import pygrank as pg
datasets = ["EUCore", "Amazon"]
pre = pg.preprocessor(assume_immutability=True, normalization="symmetric")
algs = {
"ppr.85": pg.PageRank(.85, preprocessor=pre, tol=1.E-9, max_iters=1000),
"ppr.99": pg.PageRank(.99, preprocessor=pre, tol=1.E-9, max_iters=1000),
"hk3": pg.HeatKernel(3, preprocessor=pre, tol=1.E-9, max_iters=1000),
"hk5": pg.HeatKernel(5, preprocessor=pre, tol=1.E-9, max_iters=1000),
}
algs = algs | pg.create_variations(algs, {"+Sweep": pg.Sweep})
loader = pg.load_datasets_one_community(datasets)
algs["tuned"] = pg.ParameterTuner(preprocessor=pre, tol=1.E-9, max_iters=1000)
algs["selected"] = pg.AlgorithmSelection(
pg.create_demo_filters(preprocessor=pre, tol=1.E-9,
max_iters=1000).values())
algs["tuned+Sweep"] = pg.ParameterTuner(
ranker_generator=lambda params: pg.Sweep(
pg.GenericGraphFilter(
params, preprocessor=pre, tol=1.E-9, max_iters=1000)))
for alg in algs.values():
print(alg.cite()) # prints a list of algorithm citations
pg.benchmark_print(pg.benchmark(algs, loader, pg.AUC, fraction_of_training=.5),
delimiter=" & ",
end_line="\\\\")
pg.Normalize(
postprocessor(
pg.GenericGraphFilter([1]+params,
preprocessor=pre,
error_type="iters",
max_iters=41,
optimization_dict=optimization,
preserve_norm=False))),
deviation_tol=1.E-6,
measure=measure,
optimizer=optimizer,
max_vals=[1]*40,
min_vals=[0]*40)
tuned = {
"select": pg.AlgorithmSelection(algorithms.values(), fraction_of_training=0.9, measure=measure),
"tune": create_param_tuner(),
"tuneLBFGSB": create_param_tuner(pg.lbfgsb)
}
for name, graph, group in pg.load_datasets_all_communities(datasets, min_group_size=community_size, max_group_number=3):
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=community_size, max_group_number=3)
pg.benchmark_print(
pg.benchmark_average((pg.benchmark(algorithms | tuned, loader, measure,
fraction_of_training=[0.1, 0.2, 0.3], seed=list(range(1)))), posthocs=True),
decimals=3, delimiter=" & ", end_line="\\\\")
import pygrank as pg
_, graph, communities = next(
pg.load_datasets_multiple_communities(["EUCore"], max_group_number=3))
train, test = pg.split(communities, 0.05) # 5% of community members are known
algorithms = pg.create_variations(pg.create_demo_filters(), pg.Normalize)
supervised_algorithm = pg.AlgorithmSelection(algorithms.values(),
measure=pg.AUC)
print(supervised_algorithm.cite())
modularity_algorithm = pg.AlgorithmSelection(
algorithms.values(),
fraction_of_training=1,
measure=pg.Modularity().as_supervised_method())
linkauc_algorithm = None
best_evaluation = 0
linkAUC = pg.LinkAssessment(
graph, similarity="cos",
hops=1) # LinkAUC, because emails systemically exhibit homophily
for algorithm in algorithms.values():
evaluation = linkAUC.evaluate({
community: algorithm(graph, seeds)
for community, seeds in train.items()
})
if evaluation > best_evaluation:
best_evaluation = evaluation
linkauc_algorithm = algorithm
supervised_aucs = list()
modularity_aucs = list()
linkauc_aucs = list()
preprocessor=pre,
max_iters=10000,
tol=tol,
optimization_dict=optimization),
"hk7":
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())
postprocessor = pg.Tautology
algorithms = pg.benchmarks.create_variations(algorithms, postprocessor)
measure = pg.Time
optimization = pg.SelfClearDict()
def create_param_tuner(optimizer=pg.optimize):
return pg.ParameterTuner(lambda params:
pg.Normalize(postprocessor(
pg.GenericGraphFilter([1]+params, preprocessor=pre, error_type="iters", max_iters=41,
optimization_dict=optimization, preserve_norm=False))),
deviation_tol=1.E-6, measure=measure, optimizer=optimizer, max_vals=[1]*40, min_vals=[0]*40)
tuned = {
"select": pg.AlgorithmSelection(algorithms.values()),#, combined_prediction=False),
#"tune": create_param_tuner(),
#"tuneLBFGSB": create_param_tuner(pg.lbfgsb)
}
for name, graph, group in pg.load_datasets_all_communities(datasets, min_group_size=community_size, max_group_number=3):
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=community_size, max_group_number=3)
pg.benchmark_print(
pg.benchmark_average((pg.benchmark(algorithms | tuned, loader, measure,
fraction_of_training=[0.1, 0.2, 0.3], seed=list(range(1)))), posthocs=True),
decimals=3, delimiter=" & ", end_line="\\\\")