def run_alg(alg, dsname, check_skip=True):
try:
runname = alg
if not check_skip or (check_skip and not has_run(runname, dsname)):
print("runing ", alg, dsname)
if alg == 'cdc_SVINET':
gct.run_alg(runname=runname,
data=gct.load_local_graph(dsname),
algname=alg,
seed=123,
max_iterations=1000)
else:
gct.run_alg(runname=runname,
data=gct.load_local_graph(dsname),
algname=alg,
seed=123)
print("finished", alg, dsname)
else:
print("skip", runname, dsname)
except:
print("except", runname, dsname)
fpath = os.path.join("/{}/tmp/{}_{}.done".format('data', runname, dsname))
open(fpath, 'a').close()
return True
def test_1(self):
bad_algs = [
'igraph_community_optimal_modularity', #too slow
'scan_AnyScan_ScanIdealPar', #never finish
]
runned_algs = []
algs = gct.list_algorithms()
algs = [u for u in algs if u not in bad_algs and u not in runned_algs]
datasets = [self.prefix + "_1", self.prefix + "_2"]
for dsname in datasets:
for alg in algs:
runname = alg
if not self.has_run(runname, dsname):
print("runing ", alg, dsname)
gct.run_alg(runname=runname,
data=gct.load_local_graph(dsname),
algname=alg,
seed=123)
runned_algs.append(alg)
print("AAAA", runned_algs)
results = {}
d = gct.list_all_clustering_results(print_format=False)
for u in d:
if self.prefix + "_" in u:
for run in d[u]:
a = gct.load_clustering_result(u, run)
print(u, run)
results[u + run] = a
import pandas as pd
lst = []
for alg in algs:
for dsname in datasets:
if dsname.startswith(self.prefix + '_1'):
a = []
columns = []
dsname2 = self.prefix + "_2"
a += [alg, dsname, dsname2]
columns += ['alg', 'data1', 'data2']
if dsname + alg in results and dsname2 + alg in results:
cluster1 = gct.to_cluster(results[dsname + alg])
cluster2 = gct.to_cluster(results[dsname2 + alg])
compa = gct.ClusterComparator(cluster1, cluster2)
gt = list(
gct.load_local_graph(
dsname).get_ground_truth().values())[0]
a += [
compa.sklean_nmi(), gt.is_overlap,
cluster1.is_overlap, cluster2.is_overlap,
cluster1.num_cluster, cluster2.num_cluster,
gt.num_cluster
]
columns += [
'nmi_12', 'ovp_gt', 'ovp1', 'ovp2', "#c1", "#c2",
'#c_gt'
]
compa = gct.ClusterComparator(gt, cluster1)
a += [compa.sklean_nmi()]
columns += ['nmi_t1']
compa = gct.ClusterComparator(gt, cluster2)
a += [compa.sklean_nmi()]
columns += ['nmi_t2']
lst.append(a)
else:
bad_algs.append(alg)
for alg in bad_algs:
lst.append([alg] + [None] * (len(columns) - 1))
rdf = pd.DataFrame(lst, columns=columns)
with pd.option_context('display.max_rows', 2000, 'display.max_columns',
200):
print(rdf)
rdf.to_csv(self.prefix + ".csv", index=None)
def test_1(self):
bad_algs = [
'igraph_community_optimal_modularity', #too slow
'scan_Scanpp', #never finish
] # these alg failed for this test
runned_algs = []
algs = gct.list_algorithms()
algs = [u for u in algs if u not in bad_algs and u not in runned_algs]
datasets = [self.prefix + "_w1", self.prefix + "_uw1"]
for dsname in datasets:
for alg in algs:
runname = alg
if not self.has_run(runname, dsname):
print ("runing ", alg, dsname)
if alg =='cdc_SVINET':
gct.run_alg(runname=runname, data=gct.load_local_graph(dsname), algname=alg, seed=123, max_iterations=1000)
else:
gct.run_alg(runname=runname, data=gct.load_local_graph(dsname), algname=alg, seed=123)
runned_algs.append(alg)
print ("AAAA", runned_algs)
results = {}
d = gct.list_all_clustering_results(print_format=False)
for u in d:
if self.prefix + "_" in u:
for run in d[u]:
a = gct.load_clustering_result(u, run)
print (u, run)
results[u + run] = a
import pandas as pd
lst = [];
for alg in algs:
for dsname in datasets:
if dsname.startswith(self.prefix + '_uw'):
a = [];columns = []
dsname2 = dsname.replace(self.prefix + "_uw", self.prefix + "_w")
a += [alg, dsname, dsname2]
columns += ['alg', 'data1', 'data2']
cluster1 = gct.to_cluster(results[dsname + alg])
cluster2 = gct.to_cluster(results[dsname2 + alg])
compa = gct.ClusterComparator(cluster1, cluster2)
a += [compa.sklean_nmi(), cluster1.is_overlap, cluster2.is_overlap,
cluster1.num_cluster, cluster2.num_cluster,
compa.OvpNMI()['NMImax']]
columns += ['nmi_12', 'ovp1', 'ovp2', "#c1", "#c2", 'ovpnmi_12']
gt = list(gct.load_local_graph(dsname).get_ground_truth().values())[0]
compa = gct.ClusterComparator(gt, cluster1)
a += [compa.sklean_nmi(), compa.OvpNMI()['NMImax']]
columns += ['nmi_t1', 'ovpnmi_t1']
compa = gct.ClusterComparator(gt, cluster2)
a += [compa.sklean_nmi(), compa.OvpNMI()['NMImax']]
columns += ['nmi_t2', 'ovpnmi_t2']
lst.append(a)
for alg in bad_algs:
lst.append([alg] + [None] * (len(columns) - 1))
rdf = pd.DataFrame(lst, columns=columns)
with pd.option_context('display.max_rows', 2000, 'display.max_columns', 200):
print (rdf)