def mask_data(data, mask, labels=None):
if mask is None or np.all(mask):
return data, None
else:
if labels is None:
newlabels = np.flatnonzero(mask)
else:
newlabels = labels[mask]
if data.ndim==1:
return compressed_submatrix(data, np.flatnonzero(mask)), newlabels
else:
return data[mask], newlabels
def mask_data(data, mask, labels=None):
if mask is None or np.all(mask):
return data, None
else:
if labels is None:
newlabels = np.flatnonzero(mask)
else:
newlabels = labels[mask]
if data.ndim == 1:
return compressed_submatrix(data, np.flatnonzero(mask)), newlabels
else:
return data[mask], newlabels
def Mapper_step(q, pcd, N, point_labels, filt, cover, cluster, cutoff, M,
metricpar,
verbose):
if verbose:
print ('Start Mapper thread.')
while True:
level = q.get()
if level is None: # Sentinel: end the thread
break
# Select the points in this filter range
idx = cover.data_index(level)
num_points = idx.size
# Handle special cases.
# 0 points in the filter interval: just skip the loop iteration.
if num_points == 0:
if verbose:
print('Warning! Filter level {0} is empty.'.\
format(level.index))
num_clust = 0
Z = None
R = None
# 1 point => 1 cluster
elif num_points == 1:
if verbose:
print('Warning! Filter level {0} has only one point.'.\
format(level.index))
num_clust = 1
points_clusters = np.zeros(1,dtype=int)
# We label clusters starting with 0.
Z = np.empty((0,4))
R = 0.
# 2 or more points: general case
else:
if verbose:
print('Filter level {0} has {1} points.'.\
format(level.index, num_points))
if pcd.ndim==1:
part_data = compressed_submatrix(pcd,idx)
else:
part_data = pdist(pcd[idx,:], **metricpar)
# diameter
R = part_data.max()
Z = cluster(part_data)
if Z[-1,2]>R:
print('Warning: last clustering distance is bigger than the '
'diameter of the filter slice ({0}>{1}).'.\
format(Z[-1,2], R))
R = Z[-1,2]
if cutoff:
# heights in the clustering tree
heights = Z[:,2]
# determine a cutoff value
# To do: Improve this!
num_clust = cutoff(heights, R)
# actual clustering, after the cutoff value has been determined
points_clusters = fcluster(Z, num_clust)
# My fcluster starts labelling clusters at 0!
#assert num_clust == points_clusters.max()
assert np.all(np.unique(points_clusters)==\
np.arange(num_clust))
if cutoff:
#
# Determine the nodes of the output graph
#
# Each cluster in the partial clustering gives a node
for cl in range(num_clust):
points = idx[ points_clusters == cl ] # This gives us the
# indices of the clusters points in the d matrix.
# The color is determined by the first filter component!
attribute = np.median(filt[points,0])
# Color the nodes by their median filter value
#
# Normally, the data points are labeled 0,1,...
# Allow relabeling of the data point, whatever this is good
# for.
# To do: ask Aravind.
if point_labels is not None:
points = point_labels[ points ]
M.add_node( level.index, points, attribute )
else:
# save data for the scale graph algorithm
M.scale_graph_data.append(dataidx=idx,
dendrogram=Z,
diameter=R,
levelindex=level.index)
return res
def crop(f, a, b):
from scipy.stats import scoreatpercentile
s1 = scoreatpercentile(f, a)
s2 = scoreatpercentile(f, 100-b)
assert s1<=s2
return np.logical_and(f>=s1, f<=s2)
if __name__=='__main__':
'''Test equvalence of the Python and the C++ implementation'''
import cmappertools
import numpy as np
for i in range(10000):
N = np.random.random_integers(1000)
n = np.random.random_integers(N)
dm = np.random.rand(N*(N-1)//2)
idx = np.unique(np.random.randint(N,size=n))
r = compressed_submatrix(dm,idx)
s = cmappertools.compressed_submatrix(dm,idx)
if np.any(r!=s): raise AssertionError
print("Iteration {0}: OK.".format(i))
else:
'''Load the C++ routines, if available.'''
try:
from cmappertools import compressed_submatrix
except ImportError:
sys.stderr.write("The 'cmappertools' module could not be imported.\n")
del sys
def Mapper_step(q, pcd, N, point_labels, filt, cover, cluster, cutoff, M,
metricpar, verbose):
if verbose:
print('Start Mapper thread.')
while True:
level = q.get()
if level is None: # Sentinel: end the thread
break
# Select the points in this filter range
idx = cover.data_index(level)
num_points = idx.size
# Handle special cases.
# 0 points in the filter interval: just skip the loop iteration.
if num_points == 0:
if verbose:
print('Warning! Filter level {0} is empty.'.\
format(level.index))
num_clust = 0
Z = None
R = None
# 1 point => 1 cluster
elif num_points == 1:
if verbose:
print('Warning! Filter level {0} has only one point.'.\
format(level.index))
num_clust = 1
points_clusters = np.zeros(1, dtype=int)
# We label clusters starting with 0.
Z = np.empty((0, 4))
R = 0.
# 2 or more points: general case
else:
if verbose:
print('Filter level {0} has {1} points.'.\
format(level.index, num_points))
if pcd.ndim == 1:
part_data = compressed_submatrix(pcd, idx)
else:
part_data = pdist(pcd[idx, :], **metricpar)
# diameter
R = part_data.max()
Z = cluster(part_data)
if Z[-1, 2] > R:
print('Warning: last clustering distance is bigger than the '
'diameter of the filter slice ({0}>{1}).'.\
format(Z[-1,2], R))
R = Z[-1, 2]
if cutoff:
# heights in the clustering tree
heights = Z[:, 2]
# determine a cutoff value
# To do: Improve this!
num_clust = cutoff(heights, R)
# actual clustering, after the cutoff value has been determined
points_clusters = fcluster(Z, num_clust)
# My fcluster starts labelling clusters at 0!
#assert num_clust == points_clusters.max()
assert np.all(np.unique(points_clusters)==\
np.arange(num_clust))
if cutoff:
#
# Determine the nodes of the output graph
#
# Each cluster in the partial clustering gives a node
for cl in range(num_clust):
points = idx[points_clusters == cl] # This gives us the
# indices of the clusters points in the d matrix.
# The color is determined by the first filter component!
attribute = np.median(filt[points, 0])
# Color the nodes by their median filter value
#
# Normally, the data points are labeled 0,1,...
# Allow relabeling of the data point, whatever this is good
# for.
# To do: ask Aravind.
if point_labels is not None:
points = point_labels[points]
M.add_node(level.index, points, attribute)
else:
# save data for the scale graph algorithm
M.scale_graph_data.append(dataidx=idx,
dendrogram=Z,
diameter=R,
levelindex=level.index)
def crop(f, a, b):
from scipy.stats import scoreatpercentile
s1 = scoreatpercentile(f, a)
s2 = scoreatpercentile(f, 100 - b)
assert s1 <= s2
return np.logical_and(f >= s1, f <= s2)
if __name__ == '__main__':
'''Test equvalence of the Python and the C++ implementation'''
import cmappertools
import numpy as np
for i in range(10000):
N = np.random.random_integers(1000)
n = np.random.random_integers(N)
dm = np.random.rand(N * (N - 1) // 2)
idx = np.unique(np.random.randint(N, size=n))
r = compressed_submatrix(dm, idx)
s = cmappertools.compressed_submatrix(dm, idx)
if np.any(r != s): raise AssertionError
print("Iteration {0}: OK.".format(i))
else:
'''Load the C++ routines, if available.'''
try:
from cmappertools import compressed_submatrix
except ImportError:
sys.stderr.write("The 'cmappertools' module could not be imported.\n")
del sys