def dbscan_tests(features, ftype='mean', eps=0.7, min_samples=2):
if ftype == 'min':
D = facedist.min_dist(features)
elif ftype == 'max':
D = facedist.max_dist(features)
elif ftype == 'meanmin':
D = facedist.meanmin_dist(features)
else:
D = facedist.mean_dist(features)
print D.shape
nrow = len(features)
dense_distances = np.zeros((nrow, nrow), dtype=np.double)
for ii in range(nrow):
for jj in range(ii + 1, nrow):
nn = ii + jj * (jj - 1) / 2
rd = D[nn]
dense_distances[ii, jj] = rd
dense_distances[jj, ii] = rd
del D
db = DBSCAN(eps=eps, min_samples=min_samples,
metric='precomputed').fit(dense_distances)
labels = db.labels_
unique, counts = np.unique(labels, return_counts=True)
return np.asarray((unique, counts)).T, labels, None
def hdbscan_tests(features, ftype='mean', min_cluster_size=2):
if ftype == 'min':
D = facedist.min_dist(features)
elif ftype == 'max':
D = facedist.max_dist(features)
elif ftype == 'meanmin':
D = facedist.meanmin_dist(features)
else:
D = facedist.mean_dist(features)
print D.shape
nrow = len(features)
dense_distances = np.zeros((nrow, nrow), dtype=np.double)
for ii in range(nrow):
for jj in range(ii + 1, nrow):
nn = ii + jj * (jj - 1) / 2
rd = D[nn]
dense_distances[ii, jj] = rd
dense_distances[jj, ii] = rd
del D
db = hdbscan.HDBSCAN(min_cluster_size=min_cluster_size,
metric='precomputed').fit(dense_distances)
labels = db.labels_
probabilities = db.probabilities_
pers = db.cluster_persistence_
unique, counts = np.unique(labels, return_counts=True)
return np.asarray((unique, counts)).T, labels, probabilities, pers
def dbscan_tests(features, ftype='mean', eps=0.7, min_samples=2):
if ftype=='min':
D = facedist.min_dist(features)
elif ftype=='max':
D = facedist.max_dist(features)
elif ftype=='meanmin':
D = facedist.meanmin_dist(features)
else:
D = facedist.mean_dist(features)
#print D.shape
db = DBSCAN(eps=eps, min_samples=min_samples, metric='precomputed').fit(D)
labels = db.labels_
unique, counts = np.unique(labels, return_counts=True)
return np.asarray((unique, counts)).T, labels, None
def agglo_tests(features, ftype='mean', num_cluster=8, metric='precomputed'):
if ftype=='min':
D = facedist.min_dist(features)
elif ftype=='max':
D = facedist.max_dist(features)
elif ftype=='meanmin':
D = facedist.meanmin_dist(features)
else:
D = facedist.mean_dist(features)
#print D.shape
Hclustering = AgglomerativeClustering(n_clusters=num_cluster, affinity=metric, linkage='average')
Hclustering.fit(D)
labels = Hclustering.labels_
unique, counts = np.unique(labels, return_counts=True)
return np.asarray((unique, counts)).T, labels, None
def hdbscan_tests(features, ftype='mean', min_cluster_size=2):
if ftype=='min':
D = facedist.min_dist(features)
elif ftype=='max':
D = facedist.max_dist(features)
elif ftype=='meanmin':
D = facedist.meanmin_dist(features)
else:
D = facedist.mean_dist(features)
#print D.shape
db = hdbscan.HDBSCAN(min_cluster_size=min_cluster_size, metric='precomputed').fit(D)
labels = db.labels_
probabilities = db.probabilities_
pers = db.cluster_persistence_
unique, counts = np.unique(labels, return_counts=True)
return np.asarray((unique, counts)).T, labels, probabilities, pers
#import pyximport
#pyximport.install()
import fastdist
import numpy as np
import timeit
import facedist
FA = np.zeros((90000, 100))
facedist.mean_dist(FA)
A = np.random.randn(100, 200)
F = []
for i in range(1000):
F.append(np.array([np.linspace(0.1, 1.5, 1700) for i in range(100)]))
F = np.array(F)
print F.shape
D1 = np.sqrt(np.square(A[np.newaxis, :, :] - A[:, np.newaxis, :]).sum(2))
D2 = fastdist.dist(A)
D3 = facedist.dist(F)
print np.allclose(D1, D2)
# True
#import pyximport
#pyximport.install()
import numpy as np
import timeit
import facedist
A = np.random.randn(100, 200)
F = []
for i in range(1000):
F.append(np.array([np.linspace(0.1, 1.5, 1700) for i in range(100)]))
F = np.array(F)
print F.shape
D1 = np.sqrt(np.square(A[np.newaxis, :, :] - A[:, np.newaxis, :]).sum(2))
D3 = facedist.mean_dist(F)
print np.allclose(D1, D2)
# True
#print timeit.timeit('np.sqrt(np.square(A[np.newaxis,:,:]-A[:,np.newaxis,:]).sum(2))', number=100, setup='import numpy as np; A = np.random.randn(100, 200)')
import timeit
import facedist
A = np.random.randn(100, 200)
F = []
for i in range(1000):
F.append(np.array([np.linspace(0.1,1.5,1700) for i in range(100)]))
F = np.array(F)
print F.shape
D1 = np.sqrt(np.square(A[np.newaxis,:,:]-A[:,np.newaxis,:]).sum(2))
D3 = facedist.mean_dist(F)
print np.allclose(D1, D2)
# True
#print timeit.timeit('np.sqrt(np.square(A[np.newaxis,:,:]-A[:,np.newaxis,:]).sum(2))', number=100, setup='import numpy as np; A = np.random.randn(100, 200)')
import networkx as nx
fileDir = os.path.dirname(os.path.realpath(__file__))
# Load features array from disk
features = np.load(os.path.join(fileDir, 'features_3MONTH_15.npy'))
print 'Loaded feature:', features.shape
test = features[:1000]
print test.shape
start = time.time()
D = facedist.mean_dist(test)
print 'D64 D:', D.shape
print 'D64 sys-size:', sys.getsizeof(D)
print 'D64 np nbytes:', D.nbytes
nrow = len(test)
fdense_distances = np.zeros((nrow, nrow), dtype=np.double)
for ii in range(nrow):
for jj in range(ii + 1, nrow):
nn = ii + jj * (jj - 1) / 2
rd = D[nn]
fdense_distances[ii, jj] = rd
fdense_distances[jj, ii] = rd
#import pyximport
#pyximport.install()
import fastdist
import numpy as np
import timeit
import facedist
FA = np.zeros((90000, 100))
facedist.mean_dist(FA)
A = np.random.randn(100, 200)
F = []
for i in range(1000):
F.append(np.array([np.linspace(0.1,1.5,1700) for i in range(100)]))
F = np.array(F)
print F.shape
fileDir = os.path.dirname(os.path.realpath(__file__))
# Load features array from disk
features = np.load(os.path.join(fileDir,'features_30sec_fixed.npy'))
print 'Loaded feature:', features.shape
test = features[:10000]
print test.shape
start = time.time()
D = facedist.mean_dist(test)
print 'D64 sys-size:', sys.getsizeof(D)
print 'D64 np nbytes:', D.nbytes
nrow = len(test)
dense_distances = np.zeros( (nrow, nrow), dtype=np.double)
for ii in range(nrow):
for jj in range(ii+1, nrow):
nn = ii+jj*(jj-1)/2
rd = D[nn]
dense_distances[ii, jj] = rd
dense_distances[jj, ii] = rd