def distMatrix(clusters):
"""
Calculate the all-pairs distance matrix between the given clusters
@type clusters: dict
@param clusters: The list of clusters to calculate the all-pairs distance
matrix, keyed on cluster ID. The new cluster is assumed to be the last element.
@rtype: tuple
@return: The all-pairs distance matrix and the pair with the overall minimum distance
"""
min = None; minpair = None
dist = {}
ids = np.sort(clusters.keys())
for i in ids:
dist[i] = {}
for j in ids[i+1:ids.shape[0]]:
dist[i][j] = util.nonSymmetricClusterDistance(clusters[i], clusters[j])
if (min is None) or (dist[i][j] < min):
min = dist[i][j]
minpair = (i,j)
return dist, minpair
def updateDistMatrix(clusters, matrix, remIDs, newID):
"""
Selectively update the distance calculations by deleting the removed
clusters and adding calculations for the replacement cluster
:@type clusters: dict
:@param clusters: The data split into clusters stored in numpy arrays,
keyed on ID.
:@type matrix: dict
:@param matrix: A dict containing one dict for each cluster, keyed on the
cluster id with the distance as the value.
:@type remIDs: tuple
:@param remIDs: The ids of the removed clusters.
:@type newID: int
:@param newID: The id of the replacement cluster.
"""
# remove rows
for i in remIDs:
del matrix[i]
for i in np.sort(clusters.keys()):
# remove the old cluster distance calculations (columns)
if i != newID:
for j in remIDs:
if j in matrix[i]:
del matrix[i][j]
# calculate distances to the new cluster
matrix[i][newID] = util.nonSymmetricClusterDistance(clusters[i], clusters[newID])
# calculate the minpair:
# inner min finds the minpairs for each row, outer min finds overall minpair
minpair = min([(i,min(matrix[i], key=lambda z:matrix[i].get(z))) for i in matrix],
key=lambda p: matrix.get(p[0]).get(p[1]))
# Add empty distance dict for later updates
matrix[newID] = {}
return matrix, minpair