def test_calc_best_fit_line_weighted(self):
data = np.array([[1, 3.2627812, -3.1364346],
[2, 3.4707861, -3.28776192],
[3, 3.67879099, -3.43908923]])
w = [1.0, 1.0, 1.0]
best_fit_line = utils.calc_best_fit_line_polyfit(data, w)
print best_fit_line
def calc_prep_dist(P):
prep_dist = np.full((len(P), len(P)), float("inf"))
for index, value in np.ndenumerate(prep_dist):
if index[0] < index[1]:
segment = P[index[0] : index[1] + 1, :]
best_fit_line = utils.calc_best_fit_line_polyfit(segment)
prep_dist[index] = utils.sqrd_dist_sum(segment, best_fit_line)
return prep_dist
def calc_weighted_prep_dist(pw):
prep_dist = np.full((len(pw), len(pw)), float("inf"))
for index, value in np.ndenumerate(prep_dist):
if index[0] < index[1]:
if index[1] - index[0] == 1:
prep_dist[index] = 0
continue
segment = pw[index[0] : index[1] + 1, :3]
weights = pw[index[0] : index[1] + 1, 3:].flatten()
best_fit_line = utils.calc_best_fit_line_polyfit(segment, weights)
prep_dist[index] = utils.sqrd_dist_sum_weighted(segment, best_fit_line, w=weights)
return prep_dist
def calc_coreset_prep_dist(D):
prep_dist = np.full((len(D), len(D)), float("inf"))
for (first_coreset, second_coreset), value in np.ndenumerate(prep_dist):
# we only want to calculate for segments that start in
# starting coreset endpoints and end in ending coreset endpoints
if first_coreset <= second_coreset:
C = []
for coreset in D[first_coreset:second_coreset+1]:
C.append(coreset)
coreset_of_coresets = Coreset.OneSegmentCorset(C, True)
best_fit_line = utils.calc_best_fit_line_polyfit(coreset_of_coresets.repPoints, True)
fitting_cost = utils.sqrd_dist_sum(coreset_of_coresets.repPoints, best_fit_line)*coreset_of_coresets.weight
prep_dist[first_coreset, second_coreset] = fitting_cost
return prep_dist
def calc_coreset_prep_dist(D):
prep_dist = np.full((len(D), len(D)), float("inf"))
for (first_coreset, second_coreset), value in np.ndenumerate(prep_dist):
# we only want to calculate for segments that start in
# starting coreset endpoints and end in ending coreset endpoints
if first_coreset <= second_coreset:
C = []
W = []
for coreset in D[first_coreset:second_coreset+1]:
# segment = np.vstack([segment, coreset.C.repPoints]) if segment.size else coreset.C.repPoints
C.append(coreset)
W.append(coreset.C.weight)
coreset_of_coresets = Coreset.OneSegmentCorset(C, True)
best_fit_line = utils.calc_best_fit_line_polyfit(coreset_of_coresets.repPoints, True)
# best_fit_line = utils.calc_best_fit_line(segment)
# fitting_cost = 0
# for i in xrange(len(C)):
# fitting_cost += utils.sqrd_dist_sum(C[i], best_fit_line)*W[i]
fitting_cost = utils.sqrd_dist_sum(coreset_of_coresets.repPoints, best_fit_line)*coreset_of_coresets.weight
prep_dist[first_coreset, second_coreset] = fitting_cost
return prep_dist
def BalancedPartition(P, a, bicritiriaEst, is_coreset=False):
Q = []
D = []
points = P
# add arbitrary item to list
dimensions = points[0].C.repPoints.shape[1] if is_coreset else points.shape[1]
if is_coreset:
points.append(P[0]) # arbitrary coreset n+1
else:
points = np.vstack((points, np.zeros(dimensions))) # arbitrary point n+1
n = len(points)
for i in xrange(0, n):
Q.append(points[i])
cost = one_seg_cost(np.asarray(Q), is_coreset)
# print "bp cost:", cost, "points", Q
# if current number of points can be turned into a coreset - 3 conditions :
# 1) cost passed threshold
# 2) number of points to be packaged greater than dimensions + 1
# 3) number of points left greater then dimensions + 1 (so they could be packaged later)
if cost > bicritiriaEst and (is_coreset or (len(Q) > dimensions + 1 and dimensions + 1 <= n - 1 - i)) or i == n - 1:
if is_coreset and len(Q) == 1:
if i != n - 1:
D.append(Q[0])
Q = []
continue
T = Q[:-1]
C = OneSegmentCorset(T, is_coreset)
g = utils.calc_best_fit_line_polyfit(OneSegmentCorset(np.asarray(T), is_coreset).repPoints)
if is_coreset:
b = T[0].b
e = T[-1].e
else:
b = T[0][0] # signal index of first item in T
e = T[-1][0] # signal index of last item in T
D.append(coreset(C, g, b, e))
Q = [Q[-1]]
return D