def _reduce_search_space(self, neighbors_reduction_thr=18):
"""
Wrapper function to discard clusters from the tractogram too far from
the model and logging informations.
:param neighbors_reduction_thr, float, distance in mm for thresholding
to discard distant streamlines.
"""
centroid_matrix = bundles_distances_mdf(self.model_centroids,
self.centroids)
centroid_matrix[centroid_matrix > neighbors_reduction_thr] = np.inf
mins = np.min(centroid_matrix, axis=0)
close_clusters_indices = np.array(np.where(mins != np.inf)[0],
dtype=np.int32)
self.neighb_indices = []
for i in close_clusters_indices:
self.neighb_indices.extend(self.wb_clusters_indices[i])
self.neighb_indices = np.array(self.neighb_indices, dtype=np.int32)
self.neighb_streamlines = reconstruct_streamlines_from_memmap(
self.memmap_filenames, self.neighb_indices)
self.neighb_centroids = [
self.centroids[i] for i in close_clusters_indices
]
return len(close_clusters_indices) > 0
def single_clusterize_and_rbx_init(args):
"""
Wrapper function to multiprocess clustering executions and recobundles
initialisation.
Parameters
----------
tmp_memmap_filename: tuple (3)
Temporary filename for the data, offsets and lengths.
parameters_list : tuple (3)
clustering_thr : int
Distance in mm (for QBx) to cluster the input tractogram.
seed : int
Value to initialize the RandomState of numpy.
nb_points : int
Number of points used for all resampling of streamlines.
Returns
-------
rbx : dict
Initialisation of the recobundles class using specific parameters.
"""
tmp_memmap_filename = args[0]
wb_streamlines = reconstruct_streamlines_from_memmap(tmp_memmap_filename)
clustering_thr = args[1][0]
seed = args[1][1]
nb_points = args[2]
rbx = {}
base_thresholds = [45, 35, 25]
rng = np.random.RandomState(seed)
cluster_timer = time()
# If necessary, add an extra layer (more optimal)
if clustering_thr < 15:
current_thr_list = base_thresholds + [15, clustering_thr]
else:
current_thr_list = base_thresholds + [clustering_thr]
cluster_map = qbx_and_merge(wb_streamlines,
current_thr_list,
nb_pts=nb_points,
rng=rng,
verbose=False)
clusters_indices = []
for cluster in cluster_map.clusters:
clusters_indices.append(cluster.indices)
centroids = ArraySequence(cluster_map.centroids)
clusters_indices = ArraySequence(clusters_indices)
clusters_indices._data = clusters_indices._data.astype(np.int32)
rbx[(seed, clustering_thr)] = RecobundlesX(tmp_memmap_filename,
clusters_indices,
centroids,
nb_points=nb_points,
rng=rng)
logging.info('QBx with seed {0} at {1}mm took {2}sec. gave '
'{3} centroids'.format(seed, current_thr_list,
round(time() - cluster_timer, 2),
len(cluster_map.centroids)))
return rbx