def addgraphs(graphs):
#grammar = loco.LOCO(
grammar = lsgg.lsgg(decomposition_args={
"radius_list": [0, 1, 2],
"thickness_list": [1],
"loco_minsimilarity": .8,
"thickness_loco": 4
},
filter_args={
"min_cip_count": 2,
"min_interface_count": 2
})
grammar.fit(graphs, n_jobs=args.n_jobs)
scorer = score.OneClassEstimator(n_jobs=args.n_jobs).fit(graphs)
scorer.n_jobs = 1 # demons cant spawn children
selector = choice.SelectMaxN(10)
transformer = transformutil.no_transform()
mysample = partial(sample.multi_sample,
transformer=transformer,
grammar=grammar,
scorer=scorer,
selector=selector,
n_steps=5,
n_neighbors=100)
if False:
res = sgexec.sgexec(mysample, graphs)
else:
res = ba.mpmap_prog(mysample,
graphs,
poolsize=args.n_jobs,
chunksize=1)
return graphs + res
def addgraphs(graphs):
#grammar = loco.LOCO(
grammar = lsgg.lsgg(decomposition_args={
"radius_list": [0, 1, 2],
"thickness_list": [1, 2],
"loco_minsimilarity": .8,
"thickness_loco": 4
},
filter_args={
"min_cip_count": 1,
"min_interface_count": 1
})
grammar.fit(graphs, n_jobs=args.n_jobs)
#scorer = score.OneClassEstimator(n_jobs=args.n_jobs).fit(graphs)
scorer = score.OneClassAndSizeFactor(
n_jobs=args.n_jobs,
model=svm.OneClassSVM(kernel='linear', gamma='auto')).fit(graphs)
scorer.n_jobs = 1 # demons cant spawn children
#selector = choice.SelectProbN(1)
selector = choice.SelectClassic(
reg=0) # linear kernel -> .8 , rbf kernel -> .97?
transformer = transformutil.no_transform()
# multi sample: mysample = partial(sample.multi_sample, transformer=transformer, grammar=grammar, scorer=scorer, selector=selector, n_steps=20, n_neighbors=200)
# mysample = partial(sample.sample, transformer=transformer, grammar=grammar, scorer=scorer, selector=selector, n_steps=20)
mysample = partial(sample.sample_sizeconstraint,
penalty=args.size_score_penalty,
transformer=transformer,
grammar=grammar,
scorer=scorer,
selector=selector,
n_steps=args.n_steps)
#print (mysample(graphs[0]))
#exit()
return mysample, graphs
def makelsgg():
return lsgg.lsgg(decomposition_args={
"radius_list": [0],
"thickness_list": [1]
},
filter_args={
"min_cip_count": 1,
"min_interface_count": 1
})
def __init__(self,
radius_list=[0, 1, 2, 3],
thickness_list=[1],
min_cip_count=4,
min_interface_count=2,
max_n_neighbors=None,
n_neigh_steps=1,
max_neighborhood_size=None):
"""init."""
self.max_n_neighbors = max_n_neighbors
self.n_neigh_steps = n_neigh_steps
self.max_neighborhood_size = max_neighborhood_size
self.grammar = lsgg(
decomposition_args=dict(radius_list=radius_list,
thickness_list=thickness_list,
hash_bitmask=2**16 - 1),
filter_args=dict(min_cip_count=min_cip_count,
min_interface_count=min_interface_count))
def test_dark_edges():
''''''
lsggg = lsgg.lsgg()
g = _edenize_for_testing(nx.path_graph(3))
g_dark = g.copy()
for n,d in g_dark.nodes(data=True):
if 'edge' in d:
d['nesting'] = True
# test fit
lsggg.fit([g, g_dark])
assert 2 == len([g.graph for v in lsggg.productions.values() for g in v.values()])
# test production
res = lsggg.neighbors(g_dark)
#import structout as so
#so.gprint([g.graph for v in lsggg.productions.values() for g in v.values()] )
assert 2 == len(list(res))
def __init__(self,
r=3,
d=3,
min_count=1,
max_n_neighbors=None,
n_iter=20,
k_best=5):
"""init."""
self.max_n_neighbors = max_n_neighbors
self.n_iter = n_iter
self.k_best = k_best
decomposition_args = {
"radius_list": [0, 1, 2, 3],
"thickness_list": [2]
}
filter_args = {
"min_cip_count": min_count,
"min_interface_count": min_count
}
self.grammar = lsgg(decomposition_args, filter_args)
self.multiobj_est = DistRankSizeCostEstimator(r=r, d=d)
def __init__(self,
r=3,
d=3,
min_count=1,
max_n_neighbors=None,
n_iter=20,
k_best=5,
improve=True):
"""init."""
self.r = r
self.d = d
self.max_n_neighbors = max_n_neighbors
self.n_iter = n_iter
self.k_best = k_best
self.improve = improve
decomposition_args = {
"radius_list": [0, 1, 2, 3],
"thickness_list": [2]
}
filter_args = {
"min_cip_count": min_count,
"min_interface_count": min_count
}
self.grammar = lsgg(decomposition_args, filter_args)
def test_get_grammar():
lsggg = lsgg.lsgg()
g = _edenize_for_testing(nx.path_graph(4))
lsggg.fit([g, g, g])
return lsggg