def calc_dist_matrix(self, approach=2, verbose=False):
"""
Method to calculate euclidean distance from each sample
and make a matrix
:return: 2d matrix
"""
if verbose:
Opt.cprint('Building distance matrix... ', newline='')
if approach == 1:
self.distance_matrix = np.apply_along_axis(
lambda x: Euclidean.mat_dist(x, self.matrix), 1, self.matrix)
elif approach == 2:
ndims = self.matrix.shape[1]
temp_mat = np.zeros([self.matrix.shape[0], self.matrix.shape[0]],
np.float32)
for dim in range(ndims):
arr = np.repeat(self.matrix[:, dim][:, np.newaxis], self.nsamp,
1)
arr_ = arr.T
temp_mat += (arr - arr_)**2
self.distance_matrix = np.sqrt(temp_mat)
else:
raise ValueError('Unrecognized approach')
if verbose:
Opt.cprint('Done!')
def proximity_filter(self, thresh=None, verbose=False):
"""
method to remove points based on proximity threshold
:param thresh: proximity threshold (default: 90th percentile) valid values: 1-99
:param verbose: If steps should be displayed
:return: None
"""
if verbose:
Opt.cprint('Applying proximity filter...')
if thresh is None:
thresh = self.cluster_center('percentile_90')
elif 'percentile_' in thresh:
thresh = self.cluster_center(thresh)
elif thresh in (int, float):
thresh = self.cluster_center('percentile_{}'.format(
str(int(thresh))))
else:
if verbose:
warnings.warn(
'Invalid thresh value.\n Using default: 90th percentile centroid vector.'
)
thresh = self.cluster_center('percentile_90')
# number of close proximities associated with each element
n_proxim = np.apply_along_axis(
lambda x: np.count_nonzero((x > 0.0) & (x < thresh)), 0,
self.distance_matrix)
if verbose:
Opt.cprint('Max group size : {} '.format(str(n_proxim.max())),
newline='')
Opt.cprint('Min group size : {} '.format(str(n_proxim.min())))
# sort the indices in increasing order of n_proxim
idx = []
idx += np.argsort(n_proxim).tolist()
idx_out = list()
# find indices of elements that should be removed
for ii in idx:
if ii not in idx_out:
arr = self.distance_matrix[ii, 0:(ii + 1)]
temp_list = (np.where((arr < thresh)
& (arr > 0.0))[0]).tolist()
idx_out += temp_list
idx_out = list(set(idx_out))
# sort the indices in decreasing order for pop()
pop_idx = sorted(list(set(idx_out)), reverse=True)
if verbose:
Opt.cprint('Removing {} elements...'.format(str(len(pop_idx))))
for pop_id in pop_idx:
self.samples.pop(pop_id)
self.nsamp = len(self.samples)
self.index = list(range(self.nsamp))