def bicriteria(P, k, is_coreset=False):
if len(P) <= (4 * k + 1):
return utils.best_fit_line_cost(P)
m = int(math.floor(len(P) / (4 * k)))
i = 0
j = m
# one_seg_res will hold segment starting index and result (squred distance sum)
one_seg_res = []
# partition to 4k segments and call 1-segment for each
while i < len(P):
partition_set = one_seg_cost(P[i:j], is_coreset)
one_seg_res.append((partition_set, int(i)))
i += m
j += m
# sort result
one_seg_res = sorted(one_seg_res, key=lambda res: res[0])
# res = the distances of the min k+1 segments
res = 0
# sum distances of k+1 min segments and make a list of point to delete from P to get P \ Q from the algo'
rows_to_delete = []
for i in xrange(k + 1):
res += one_seg_res[i][0]
for j in xrange(m):
rows_to_delete.append(one_seg_res[i][1] + j)
P = np.delete(P, rows_to_delete, axis=0)
return res + bicriteria(P, k, is_coreset)
def one_seg_cost(P, is_coreset=False):
if is_coreset:
oneSegmentCoreset = OneSegmentCorset(P, is_coreset)
return utils.best_fit_line_cost(oneSegmentCoreset.repPoints, is_coreset) * oneSegmentCoreset.weight
else:
return utils.best_fit_line_cost(P, is_coreset)