def cluster_experiments(nets, y, dataset, n_runs, test_run):
max_size = int(np.floor(np.log2(len(nets.nodes()))))
sizes = [2**i for i in range(1, max_size + 1)]
parts = create_metis_partitions([nets], sizes=sizes, seed=123)
parts = parts[0]
results = []
ohe = OneHotEncoder(n_values='auto', categorical_features='all')
y = y.loc[parts.index] # remove singletones
for n_cluster in sizes:
X = parts[str(n_cluster) + '_' + dataset].values + 1
X = ohe.fit_transform(np.expand_dims(X, axis=1))
df = evaluate_model(X, y, n_runs=n_runs, test_run=test_run)
df['n_cluster'] = n_cluster
results.append(df)
# use all partiionings
X = parts.values + 1
X = ohe.fit_transform(X)
df = evaluate_model(X, y, n_runs=n_runs, test_run=test_run)
df['n_cluster'] = 0
results.append(df)
results = pd.concat(results)
if not os.path.exists('../results/' + dataset + '/'):
os.makedirs('../results/' + dataset + '/')
results.to_csv('../results/' + dataset + '/cluster', index=False)
return results
def bow_class_count_experiments(G, X, y, dataset, max_k, n_runs, test_run):
from feature_extract import get_neighbor_ids
adjs = get_neighbor_ids(G, y.index.values, max_k=max_k)
results = []
# bow and counts
print 'bow + counts'
results = evaluate_model(X, y, adjs=adjs, n_runs=n_runs,
test_run=test_run)
if not os.path.exists('../results/' + dataset + '/'):
os.makedirs('../results/' + dataset + '/')
results.to_csv('../results/' + dataset + '/bow_counts', index=False)
# bow
print 'bow'
results = evaluate_model(X, y, adjs=None, n_runs=n_runs,
test_run=test_run)
results.to_csv('../results/' + dataset + '/bow', index=False)
# counts
print 'counts'
results = evaluate_model(None, y, adjs=adjs, n_runs=n_runs,
test_run=test_run)
results.to_csv('../results/' + dataset + '/counts', index=False)
return results
def id_cluster_experiments(nets, y, dataset, n_runs, test_run):
from feature_extract import get_neighbor_ids
adjs = get_neighbor_ids(nets, y.index.values, max_k=1)
max_size = int(np.floor(np.log2(len(nets.nodes()))))
sizes = [2**i for i in range(1, max_size + 1)]
parts = create_metis_partitions([nets], sizes=sizes, seed=123)
parts = parts[0]
results = []
ohe = OneHotEncoder(n_values='auto', categorical_features='all')
y = y.loc[parts.index] # remove singletones
# use all partiionings
X = parts.values + 1
X = ohe.fit_transform(X)
X = sp.hstack([X, adjs[0]])
results = evaluate_model(X, y, n_runs=n_runs, test_run=test_run)
if not os.path.exists('../results/' + dataset + '/'):
os.makedirs('../results/' + dataset + '/')
results.to_csv('../results/' + dataset + '/id_cluster', index=False)
return results
def class_count_experiments(G,
y,
dataset,
max_k,
n_runs,
test_run,
ignore_unlabeled=False):
from feature_extract import get_neighbor_ids
adjs = get_neighbor_ids(G, y.index.values, max_k=max_k)
results = []
for depth in range(max_k):
print "depth", depth
df = evaluate_model(None,
y,
adjs=adjs[:depth + 1],
n_runs=n_runs,
test_run=test_run,
ignore_unlabeled=ignore_unlabeled)
df['depth'] = depth
results.append(df)
results = pd.concat(results)
if not os.path.exists('../results/' + dataset + '/'):
os.makedirs('../results/' + dataset + '/')
if ignore_unlabeled:
results.to_csv('../results/' + dataset + '/ids_labeled', index=False)
else:
results.to_csv('../results/' + dataset + '/counts', index=False)
return results
def id_class_count_experiments(G, y, dataset, max_k, n_runs, test_run):
from feature_extract import get_neighbor_ids
adjs = get_neighbor_ids(G, y.index.values, max_k=max_k)
results = []
X = adjs[0].todense()
results = evaluate_model(X, y, adjs=adjs, n_runs=n_runs, test_run=test_run)
if not os.path.exists('../results/' + dataset + '/'):
os.makedirs('../results/' + dataset + '/')
results.to_csv('../results/' + dataset + '/id_counts', index=False)
return results
def id_rwr_experiments(G, y, dataset, n_runs, test_run):
from feature_extract import get_neighbor_ids, calc_rwr_matrix
adjs = get_neighbor_ids(G, y.index.values, max_k=1)
X = []
X = calc_rwr_matrix(np.array(adjs[0].todense()), 0.9)
X = np.hstack([X, adjs[0].todense()])
results = evaluate_model(X, y, n_runs=n_runs, test_run=test_run)
if not os.path.exists('../results/' + dataset + '/'):
os.makedirs('../results/' + dataset + '/')
results.to_csv('../results/' + dataset + '/id_rwr', index=False)
return results
def ids_experiment(G, y, dataset, max_k, n_runs, test_run):
from feature_extract import get_neighbor_ids
adjs = get_neighbor_ids(G, y.index.values, max_k=max_k)
results = []
X = []
for depth in range(max_k):
X = sp.hstack(adjs[:depth + 1])
df = evaluate_model(X, y, n_runs=n_runs, test_run=test_run)
df['depth'] = depth
results.append(df)
results = pd.concat(results)
if not os.path.exists('../results/' + dataset + '/'):
os.makedirs('../results/' + dataset + '/')
results.to_csv('../results/' + dataset + '/ids', index=False)
return results
def class_prob_experiments(G, y, dataset, max_k, n_runs, test_run):
from feature_extract import get_neighbor_ids
adjs = get_neighbor_ids(G, y.index.values, max_k=max_k)
results = []
for depth in range(max_k):
df = evaluate_model(None,
y,
adjs=adjs[:depth + 1],
normelize=True,
n_runs=n_runs,
test_run=test_run)
df['depth'] = depth
results.append(df)
results = pd.concat(results)
if not os.path.exists('../results/' + dataset + '/'):
os.makedirs('../results/' + dataset + '/')
results.to_csv('../results/' + dataset + '/proba', index=False)
return results
def rwr_experiments(G, y, dataset, n_runs, test_run):
from feature_extract import get_neighbor_ids, calc_rwr_matrix
adjs = get_neighbor_ids(G, y.index.values, max_k=1)
results = []
X = []
c_range = np.arange(0.1, 1, 0.2)
if test_run:
c_range = [0.9]
for c in c_range:
X = calc_rwr_matrix(np.array(adjs[0].todense()), c)
df = evaluate_model(X, y, n_runs=n_runs, test_run=test_run)
df['restart'] = c
results.append(df)
results = pd.concat(results)
if not os.path.exists('../results/' + dataset + '/'):
os.makedirs('../results/' + dataset + '/')
results.to_csv('../results/' + dataset + '/rwr', index=False)
return results