def fast_clustering(coords):
MAX_RMSD = 30 #Max RMSD to be considered in a cluster.
for MAX_RMSD in [
10, 15, 20, 25, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 30
]:
clusters = []
i_to_clust = []
for i, coord in enumerate(coords):
for j, cluster in enumerate(clusters):
if ftur.rmsd(coord, coords[cluster[0]]) < MAX_RMSD:
cluster.append(i)
i_to_clust.append(j)
break
else:
clusters.append([i])
i_to_clust.append(len(clusters) - 1)
print("{} maxRMSD, {} clusters".format(MAX_RMSD, len(clusters)))
print("Starting dists")
dists = np.zeros((len(coords), len(coords)))
for i, j in it.combinations(range(len(coords)), 2):
if i_to_clust[i] == i_to_clust[j]:
dists[i, j] = ftur.rmsd(coords[i], coords[j])
else:
dists[i, j] = float('inf')
return i_to_clust, dists
def topN_closest_cluster_nodes(self, n, img_index):
query_cluster = self.labels[img_index]
i = 0
cluster = []
for label in self.labels:
if label == query_cluster and i != img_index:
cluster.append(i)
i += 1
dist = dict(
map(
lambda x:
(x, distance.euclidean(self.data[img_index], self.data[x])),
cluster))
tops = sorted(dist.items(), key=itemgetter(1))[:n]
return tops
def outwrite(vocab,labels, outpath):
outfile = open(outpath,'w')
labels1 = []
for i in labels:
try:
labels1.append(i[0])
except:
labels1.append(i)
for i in set(labels1):
cluster = []
str1 = ''
for (word,label) in itertools.izip(vocab,labels):
if label==i:
cluster.append(word)
str1 = str1 + word + ' '
outfile.write("Cluster "+str(i)+":\t" + str1+'\n')
def find_boundarieskmeans(binarylist, numclusters, clusterimportance):
'''
:param binarylist: list of clustering labels
:param numclusters: number of clusters to identify
:param clusterimportance: mu as discussed in thesis, normally expected segmentsize*0.5
:return:boundaries detected using kboundaries
'''
def customdifference(point1, point2):
if point1[0] == point2[0]:
modifier = 0
else:
modifier = clusterimportance
return abs(point1[1] - point2[1]) + modifier
size = len(binarylist) / numclusters
centers = [
(i, size * i + size / 2) for i in range(numclusters)
] #Initial guess is evenly spread centroids. This should conform to our expectation
for i in range(25):
clusters = [([], centers[i]) for i in range(numclusters)]
for j in range(len(binarylist)):
value = binarylist[j]
point = [value, j]
bestcenter = min(
centers,
key=lambda x: customdifference(point, x)) #Allocation step
for cluster, center in clusters:
if center == bestcenter:
cluster.append(point)
#Updating center
centers = []
for cluster, center in clusters:
if len(cluster) > 0:
clustervalue = np.mean([x[0] for x in cluster])
numericcenter = np.mean([x[1] for x in cluster])
else:
clustervalue = int(round(random.random() * numclusters))
numericcenter = random.random() * len(binarylist)
centers.append([int(round(clustervalue)), numericcenter])
#Here we get boundaries from the centers
boundaries = boundariesfromcenters(clusters)
return boundaries
def fast_clustering(coords):
MAX_RMSD=30 #Max RMSD to be considered in a cluster.
for MAX_RMSD in [10,15,20,25,35,40,45,50,55,60,65,70,75,80,30]:
clusters = []
i_to_clust=[]
for i, coord in enumerate(coords):
for j, cluster in enumerate(clusters):
if ftur.rmsd(coord, coords[cluster[0]])<MAX_RMSD:
cluster.append(i)
i_to_clust.append(j)
break
else:
clusters.append([i])
i_to_clust.append(len(clusters)-1)
print("{} maxRMSD, {} clusters".format(MAX_RMSD, len(clusters)))
print("Starting dists")
dists = np.zeros((len(coords), len(coords)))
for i,j in it.combinations(range(len(coords)),2):
if i_to_clust[i]==i_to_clust[j]:
dists[i,j]=ftur.rmsd(coords[i], coords[j])
else:
dists[i,j]= float('inf')
return i_to_clust, dists
def filter_isolated_idxs_new(idxs=[], maxdist=3.0):
clusters = []
cluster = []
newidxs = np.zeros_like(idxs)
for i in xrange(len(idxs)):
if idxs[i]:
cluster.append(i)
if (i == len(idxs) or not idxs[i]) and len(cluster):
clusters.append(cluster)
cluster=[]
for i in xrange(len(clusters)):
valid = True
if len(clusters[i]) <= 1:
valid = False
if valid:
newidxs[clusters[i]] = True
#print clusters[i], m.lcs[0][clusters[i]]
return newidxs
def _expand_cluster(self, sample_i, neighbors):
cluster = [sample_i]
# Iterate through neighbors
for neighbor_i in neighbors:
if not neighbor_i in self.visited_samples:
self.visited_samples.append(neighbor_i)
# Fetch the samples distant neighbors
self.neighbors[neighbor_i] = self._get_neighbors(neighbor_i)
# Make sure the neighbors neighbors are more than min_samples
if len(self.neighbors[neighbor_i]) >= self.min_samples:
# Choose neighbors of neighbor except for sample
distant_neighbors = self.neighbors[neighbor_i][np.where(
self.neighbors[neighbor_i] != sample_i)]
# Add the neighbors neighbors as neighbors of sample
self.neighbors[sample_i] = np.concatenate(
(self.neighbors[sample_i], distant_neighbors))
# Expand the cluster from the neighbor
expanded_cluster = self._expand_cluster(
neighbor_i, self.neighbors[neighbor_i])
# Add expanded cluster to this cluster
cluster = cluster + expanded_cluster
if not neighbor_i in np.array(self.clusters):
cluster.append(neighbor_i)
return cluster
# Welcher Datensatz das Cluster-Zentrum bildet
centers_indices = aff_prop_cluster.cluster_centers_indices_
print "--labels_---------------"
print [ lab for lab in labels ]
print "-- cluster_centers_indices_ -------"
print [i for i in centers_indices]
cluster = []
for c in centers_indices:
cluster.append([])
for i, label in enumerate(labels):
#print names[i]," : ", label
cluster[label].append( [ i, symbols[i], names[i]] )
print "cluster"
print cluster
for j, cl in enumerate(cluster):
# Bessere Ansicht
# plt.figure()