def dbscan_score_points(self):
"""
Scoring function for the dbscan method
:return: A list of scores corresponding to the points
"""
positive_assignment = learning.dbscan(self.positive_data, self.eps[0], self.min_samples[0])
negative_assignment = learning.dbscan(self.negative_data, self.eps[1], self.min_samples[1])
# resort to k-means if things go poorly...
if max(positive_assignment) < 2:
logger.warning("Clustering positive with k-means instead...")
positive_assignment = learning.kmeans(self.positive_data, self.k_clusters_positive)
if max(negative_assignment) < 2:
logger.warning("Clustering negative with k-means instead...")
negative_assignment = learning.kmeans(self.negative_data, self.k_clusters_negative)
positive_centroids = learning.get_centroids(self.positive_data, positive_assignment)
negative_centroids = learning.get_centroids(self.negative_data, negative_assignment)
scores = [0] * self.num_points
for i in xrange(self.num_points):
point = self.data_points[i]
closest_positive = learning.closest_to(point, positive_centroids)
closest_negative = learning.closest_to(point, negative_centroids)
scores[i] = [self.proximity_metric(point, closest_positive, closest_negative)]
return scores
def silhouette_score_points(self):
"""
Scoring function for the silhouette method
:return: A list of scores corresponding to the points
"""
positive_appended = np.append(self.positive_data, self.data_points, axis=0)
negative_appended = np.append(self.positive_data, self.data_points, axis=0)
positive_assignment = learning.kmeans(positive_appended, 86)
negative_assignment = learning.kmeans(negative_appended, 86)
pos_sils = learning.silhouettes(positive_appended, positive_assignment)
neg_sils = learning.silhouettes(negative_appended, negative_assignment)
scores = np.array(pos_sils[-self.num_points:] - neg_sils[-self.num_points:])
return scores
def kmeans_score_points(self):
"""
Scoring function for the kmeans method
:return: A list of scores corresponding to the points
"""
positive_assignment = learning.kmeans(self.positive_data, self.k_clusters)
negative_assignment = learning.kmeans(self.negative_data, self.k_clusters)
positive_centroids = learning.get_centroids(self.positive_data, positive_assignment)
negative_centroids = learning.get_centroids(self.negative_data, negative_assignment)
scores = np.zeros(self.num_points)
for i in xrange(self.num_points):
point = self.data_points[i]
closest_positive = learning.closest_to(point, positive_centroids)
closest_negative = learning.closest_to(point, negative_centroids)
scores[i] = self.proximity_metric(point, closest_positive, closest_negative)
return scores
def get_assignment(self):
"""
This function is for getting the cluster assignment of the data
:return: Whatever data is returned by the dbscan wrapper function in learning.py
"""
if self.dbscan:
if self.cluster_on_tsne:
self.assignment = learning.dbscan(
self.tsne_data,
eps=self.eps,
min_samples=self.min_samples,
sort_by_size=self.order_clusters_by_size)
else:
self.assignment = learning.dbscan(
self.features,
eps=self.eps,
min_samples=self.min_samples,
sort_by_size=self.order_clusters_by_size)
elif self.kmeans:
if self.cluster_on_tsne:
self.assignment = learning.kmeans(
self.tsne_data,
self.k_clusters,
sort_by_size=self.order_clusters_by_size)
else:
self.assignment = learning.kmeans(
self.features,
self.k_clusters,
sort_by_size=self.order_clusters_by_size)
else:
self.assignment = None
self.num_clusters = len(set(self.assignment) - set([-1]))
if self.num_clusters > 0:
logger.debug("Number of clusters: %d" % self.num_clusters)
else:
logger.warning("Data was assigned to zero clusters. Exiting.")
exit()
return self.assignment
ids, data = fileIO.read_feature_file(args.features_file, normalize=True)
k_clusters = np.arange(10, 600, 10)
k_clusters = np.array(sorted(list(set(k_clusters))))
sil_scores = np.zeros(k_clusters.shape)
sil_score_std = np.zeros(k_clusters.shape)
num_repeats = 5
for i in xrange(k_clusters.shape[0]):
k = k_clusters[i]
means = np.zeros(num_repeats)
for j in xrange(num_repeats):
sils = learning.silhouette_score(
data, learning.kmeans(data, k, verbose=True))
means[j] = np.mean(sils)
sil_scores[i] = np.mean(means)
sil_score_std[i] = np.std(means)
fig = plt.figure()
ax = plt.subplot(111)
ax.plot(k_clusters, sil_scores)
ax.errorbar(k_clusters, sil_scores, yerr=sil_score_std)
ax.set_xlabel("Number of Clusters")
ax.set_ylabel("Silhouette")
ax.set_title(args.features_file)
if args.output_file is None:
filename = "%s_sil.svg" % os.path.splitext(