def fit_predict(self, data):
graph = induce_graph(data, distance=self.metric)
fname = save_to_file(graph, force_integer=True)
output_directory = '_labellings'
if not exists(output_directory):
mkdirs(output_directory)
cwd = os.getcwd()
os.chdir('YWWTools/target')
result_blocks = []
for num_blocks in range(2, 15):
result_fname = '../../%s_blocks_%d' % (fname, num_blocks)
call(' '.join([
'java',
'-cp YWWTools.jar:deps.jar yang.weiwei.Tools',
'--tool wsbm',
'--nodes %d' % graph.num_vertices(),
'--blocks %d' % num_blocks,
'--graph ../../%s' % fname,
'--output %s' % result_fname,
'--no-verbose',
]),
shell=True)
blocks = np.genfromtxt(result_fname)
score = silhouette_score(data, blocks)
result_blocks.append((score, blocks))
os.remove(result_fname)
os.chdir(cwd)
os.remove(fname)
return np.array(max(result_blocks)[1])
def plot_blocks_wsbm_evaluation_metrics(dataset,
test_range=tuple(range(2, 15))):
"""This plots the difference between the silhouette score and NMI on a
dataset when clustering with the WSBM."""
dataset_ = Table(dataset)
data, y = dataset_.X, dataset_.Y
graph = induce_graph(data)
fname = save_to_file(graph, force_integer=True)
output_directory = '_labellings'
if not exists(output_directory):
mkdirs(output_directory)
cwd = os.getcwd()
os.chdir('YWWTools/target')
nmi_scores, silhouettes_scores = [], []
for num_blocks in test_range:
result_fname = '../../%s_blocks_%d' % (fname, num_blocks)
call(' '.join([
'java',
'-cp YWWTools.jar:deps.jar yang.weiwei.Tools',
'--tool wsbm',
'--nodes %d' % graph.num_vertices(),
'--blocks %d' % num_blocks,
'--graph ../../%s' % fname,
'--output %s' % result_fname,
'--no-verbose',
]),
shell=True)
blocks = np.genfromtxt(result_fname)
os.remove(result_fname)
silhouettes_scores.append(silhouette_score(data, blocks))
nmi_scores.append(normalized_mutual_info_score(y, blocks))
os.chdir(cwd)
os.remove(fname)
plt.plot(test_range, nmi_scores, label='NMI')
plt.plot(test_range, silhouettes_scores, label='Silhouette')
plt.title('Clustering with WSBM on %s' % dataset.title())
plt.xlabel('Blocks')
plt.ylabel('Scores')
plt.legend()
plt.savefig('%s/wsbm_blocks_%s.png' % (RESULTS_DIR, dataset))
def fit_predict(self, data):
graph = induce_graph(data, distance=self.metric)
result_blocks = []
weights = graph.edge_properties['weights'].get_array()
for threshold in np.linspace(0, weights.max(), self.cutoff_interval):
working_graph = cutoff(graph, threshold, inplace=True)
# Apply the sbm to the pruned graph
blocks = minimize_blockmodel_dl(working_graph)
blocks = blocks.get_blocks().get_array()
# Silhouette doesn't work if there's only one cluster label
if len(np.unique(blocks)) > 1:
cutoff_score = silhouette_score(data, blocks)
result_blocks.append((cutoff_score, blocks))
return np.array(max(result_blocks)[1])
def plot_threshold_sbm_distance_metrics(dataset, split_into=20):
dataset_ = Table(dataset)
data, y = dataset_.X, dataset_.Y
for metric in DISTANCES:
graph = induce_graph(data, distance='%s_invexp' % metric)
weights = graph.edge_properties['weights'].get_array()
silhouette_scores = []
thresholds = np.linspace(0, weights.max(), split_into)
for threshold in thresholds:
_, silhouette = sbm_clustering_nmi_silhouette(data, y, threshold)
silhouette_scores.append(silhouette)
plt.plot(thresholds, silhouette_scores, label=metric)
plt.title('Clustering with SBM after thresholding on %s' %
dataset.title())
plt.xlabel('Distance threshold')
plt.ylabel('Silhouette score')
plt.legend()
plt.savefig('%s/threshold_clustering_metrics_%s.png' %
(RESULTS_DIR, dataset))
def plot_threshold_sbm_components(dataset, split_into=10):
"""This plots the difference between the silhouette score and NMI on a
dataset when clustering with the threshold SBM."""
dataset_ = Table(dataset)
data, y = dataset_.X, dataset_.Y
graph = induce_graph(data)
weights = graph.edge_properties['weights'].get_array()
nmi_scores, silhouettes_scores = [], []
thresholds = np.linspace(0, weights.max(), split_into)
for threshold in thresholds:
nmi, silhouette = sbm_clustering_nmi_silhouette(data, y, threshold)
nmi_scores.append(nmi)
silhouettes_scores.append(silhouette)
plt.plot(thresholds, nmi_scores, label='NMI')
plt.plot(thresholds, silhouettes_scores, label='Silhouette')
plt.title('Clustering with SBM with thresholding on %s' % dataset.title())
plt.xlabel('Distance threshold')
plt.ylabel('Scores')
plt.legend()
plt.savefig('%s/threshold_clustering_%s.png' % (RESULTS_DIR, dataset))
import matplotlib
from Orange.data import Table
matplotlib.use('Agg')
from graph_tool.draw import sfdp_layout, graph_draw
from induce_graph import induce_graph, cutoff
iris = Table('iris')
graph = induce_graph(iris)
graph_1 = cutoff(graph, 0.1)
graph_2 = cutoff(graph, 0.2)
graph_3 = cutoff(graph, 0.3)
vertex_layout = sfdp_layout(graph_2)
rgb_colors = ([[70, 190, 250]] * 50) + ([[237, 70, 47]] *
50) + ([[170, 242, 43]] * 50)
rgb_colors = [[r / 255, g / 255, b / 255] for r, g, b in rgb_colors]
colors = graph_1.new_vertex_property('vector<double>')
for node, color in zip(graph_1.vertices(), rgb_colors):
colors[node] = color
graph_draw(graph_1,
vertex_layout,
vertex_fill_color=colors,
output='iris_threshold_01.png',
output_size=(600, 400))