def select(self):
shape = self.domain_shape
if shape == (1, 1):
# skip the calucation of newgrids if shape is of size 1
matrix = sparse.csr_matrix(([1], ([0], [0])), shape=(1, 1))
newgrid = 1
else:
eps = self.eps_par
cur_noisy_x = self.x_hat
noisycnt = cur_noisy_x.sum()
# compute second level grid size
if noisycnt <= 0:
m2 = 1
else:
m2 = int(math.sqrt(noisycnt * eps / self.c2) - 1) + 1
M2 = m2**2
nn, mm = shape
newgrid = int(math.sqrt(nn * mm * 1.0 / M2) - 1) + 1
if newgrid <= 0:
newgrid = 1
num1 = int(util.old_div((nn - 1), newgrid) + 1)
num2 = int(util.old_div((mm - 1), newgrid) + 1)
# generate cell and pending queries base on new celss
cells = GenerateCells(nn, mm, num1, num2, newgrid)
matrix = cells_to_query(cells, (nn, mm))
return matrix
def mapping(self):
n, m = self.domain_shape
N = self.data_sum
eps = self.eps_par
if self.ag_flag:
m1 = int(math.sqrt((N * eps) / self.c) / 4 - 1) + 1
if m1 < 10:
m1 = 10
M = m1**2
grid = int(math.sqrt(n * m * 1.0 / M) - 1) + 1
if grid <= 0:
grid = 1
else:
M = util.old_div((N * eps), self.c)
if self.gz == 0:
grid = int(math.sqrt(n * m / M) - 1) + 1
else:
grid = int(self.gz)
if grid < 1:
grid = 1
num1 = int(util.old_div((n - 1), grid) + 1)
num2 = int(util.old_div((m - 1), grid) + 1)
# TODO: potential optimization if grid ==1 identity workload
cells = UGridPartition.GenerateCells(n, m, num1, num2, grid)
return cells_to_mapping(cells, (n, m))
def select(self):
shape = self.domain_shape
if shape == (1, 1):
# skip the calucation of newgrids if shape is of size 1
return workload.RangeQueries((1, 1),
lower=np.array([[0, 0]]),
higher=np.array([[0, 0]]))
else:
eps = self.eps_par
cur_noisy_x = self.x_hat
noisycnt = cur_noisy_x.sum()
# compute second level grid size
if noisycnt <= 0:
m2 = 1
else:
m2 = int(math.sqrt(noisycnt * eps / self.c2) - 1) + 1
M2 = m2**2
nn, mm = shape
newgrid = int(math.sqrt(nn * mm * 1.0 / M2) - 1) + 1
if newgrid <= 0:
newgrid = 1
num1 = int(util.old_div((nn - 1), newgrid) + 1)
num2 = int(util.old_div((mm - 1), newgrid) + 1)
# generate cell and pending queries base on new celss
lower, higher = AdaptiveGrid.grid_split_range(
(0, 0), (nn - 1, mm - 1), branching_list=[num1, num2])
return workload.RangeQueries(self.domain_shape, np.array(lower),
np.array(higher))
def rect_to_quads(x):
'''
Given an np array it splits it correctly to 4 quads in the midpoints
can handle arrays of arbitrary shape (1D as well)
'''
n_rows = x.shape[0]
n_cols = x.shape[1]
# If ncol is odd, do vert splits in balanced manner
col_parity = 0
if n_cols % 2:
col_parity = 1
col_midpoint = util.old_div(x.shape[1], 2)
row_midpoint = util.old_div(x.shape[0], 2)
if x.shape[0] == 1:
# if x has only one row then do only vertical split
x1, x2 = np.split(x, [col_midpoint], axis=1)
return [x1, x2]
if x.shape[1] == 1:
# if x has only one col then do only horizontal split
x1, x2 = np.split(x, [row_midpoint], axis=0)
return [x1, x2]
# o/w do both splits
x_h1, x_h2 = np.split(x, [row_midpoint], axis=0)
x1, x2 = np.split(x_h1, [col_midpoint], axis=1)
x3, x4 = np.split(x_h2, [col_midpoint + col_parity], axis=1)
return [x1, x2, x3, x4]
def hilbert(N):
"""
Produce coordinates of an NxN Hilbert curve.
@param N:
the length of side, assumed to be a power of 2 ( >= 2)
@returns:
x and y, each as an array of integers representing coordinates
of points along the Hilbert curve. Calling plot(x, y)
will plot the Hilbert curve.
From Wikipedia
"""
assert 2**int(math.ceil(math.log(
N, 2))) == N, "N={0} is not a power of 2!".format(N)
if N == 2:
return np.array((0, 0, 1, 1)), np.array((0, 1, 1, 0))
else:
x, y = HilbertTransform.hilbert(util.old_div(N, 2))
xl = np.r_[y, x, util.old_div(N, 2) + x, N - 1 - y]
yl = np.r_[x,
util.old_div(N, 2) + y,
util.old_div(N, 2) + y,
util.old_div(N, 2) - 1 - x]
return xl, yl
def select(self):
n, m = self.domain_shape
N = self.data_sum
eps = self.eps_par
if self.ag_flag:
m1 = int(math.sqrt((N * eps) / self.c) / 4 - 1) + 1
if m1 < 10:
m1 = 10
M = m1**2
grid = int(math.sqrt(n * m * 1.0 / M) - 1) + 1
if grid <= 0:
grid = 1
else:
M = util.old_div((N * eps), self.c)
if self.gz == 0:
grid = int(math.sqrt(n * m / M) - 1) + 1
else:
grid = int(self.gz)
if grid < 1:
grid = 1
num1 = int(util.old_div((n - 1), grid) + 1)
num2 = int(util.old_div((m - 1), grid) + 1)
lower, upper = GenerateCells(n, m, num1, num2, grid)
return workload.RangeQueries((n, m), np.array(lower), np.array(upper))
def Run(self, QtQ, x, epsilon, seed):
"""
QtQ - given the workload Q in matrix form, QtQ is the
multiplication between the transpose of Q and Q.
"""
assert seed is not None, 'seed must be set'
prng = numpy.random.RandomState(seed)
x = numpy.array(x)
assert len(
x.shape
) == 1, '%s is defined for 1D data only' % self.__class__.__name__
n = len(x)
err, inv, dist, query = self._GreedyHierByLv(QtQ, n, 0, withRoot=False)
qmat = []
y2 = []
for c in range(len(dist)):
if dist[c] > 0:
lb, rb = query[c]
currow = numpy.zeros(n)
currow[lb:rb + 1] = dist[c]
qmat.append(currow)
y2.append(sum(x[lb:(rb + 1)]) * dist[c])
qmat = numpy.array(qmat)
y2 += prng.laplace(0.0, util.old_div(1.0, epsilon), len(y2))
return numpy.dot(inv, numpy.dot(qmat.T, y2))
def setUp(self):
n = 1024
scale = 1E5
self.hist = numpy.array(list(range(n)))
self.d = dataset.Dataset(self.hist, None)
self.dist = numpy.random.exponential(1, n)
self.dist = util.old_div(self.dist, float(self.dist.sum()))
self.ds = dataset.DatasetSampled(self.dist, scale, None, 1001)
def GenerateCells(n, m, num1, num2, grid):
# this function used to generate all the cells in UGrid
assert math.ceil(util.old_div(n, float(grid))) == num1 and math.ceil(
util.old_div(m, float(grid))
) == num2, "Unable to generate cells for Ugrid: check grid number and grid size"
cells = []
for i in range(num1):
for j in range(num2):
lb = [int(i * grid), int(j * grid)]
rb = [int((i + 1) * grid - 1), int((j + 1) * grid - 1)]
if rb[0] >= n:
rb[0] = int(n - 1)
if rb[1] >= m:
rb[1] = int(m - 1)
cells = cells + [[lb, rb]]
return cells
def _rebuild(partition, counts, n):
"""Rebuild an estimated data using uniform expansion."""
estx = numpy.zeros(n)
n2 = len(counts)
for c in range(n2):
lb, rb = partition[c]
estx[lb:(rb + 1)] = util.old_div(counts[c], float(rb - lb + 1))
return estx
def get_A(M, noise_scales):
"""
Calculate matrix 'A' of measurements, scaled appropriately for inference
"""
sf = (util.old_div(1.0, np.array(noise_scales))
) # reciprocal of each noise scale
D = sparse.spdiags(sf, 0, sf.size, sf.size)
return D * M # scale rows
def get_y(ans, noise_scales):
"""
Calculate 'y' of answers, scaled appropriately for inference
"""
sf = (util.old_div(1.0, np.array(noise_scales))
) # reciprocal of each noise scale
y = ans * sf # element-wise multiplication
y = y[:, np.newaxis] # make column vector
return y
def L1partition_approx(x, epsilon, ratio=0.5, gethist=False,seed =None):
"""Compute the noisy L1 histogram using interval buckets of size 2^k
Args:
x - list of numeric values. The input data vector
epsilon - double. Total private budget
ratio - double in (0, 1) the use ratio*epsilon for partition computation and (1-ratio)*epsilon for querying
the count in each partition
gethist - boolean. If set to truth, return the partition directly (the privacy budget used is still ratio*epsilon)
Return:
if gethist == False, return an estimated data vector. Otherwise, return the partition
"""
assert seed is not None, "seed must be set"
prng = numpy.random.RandomState(seed)
n = len(x)
# check that the input vector x is of appropriate type
assert (x.dtype == numpy.dtype(int) or x.dtype == numpy.dtype("int32")), "Input vector must be int! %s given" %x.dtype
y=x.astype('int32') #numpy type int32 is not not JSON serializable
check = (x ==y)
assert check.sum() == len(check), "Casting error from int to int32"
x=y
hist = cutil.L1partition_approx(n+1, x, epsilon, ratio, prng.randint(500000))
hatx = numpy.zeros(n)
rb = n
if gethist:
bucks = []
for lb in hist[1:]:
bucks.insert(0, [lb, rb-1])
rb = lb
if lb == 0:
break
return bucks
else:
for lb in hist[1:]:
hatx[lb:rb] = util.old_div(max(0, sum(x[lb:rb]) + prng.laplace(0, util.old_div(1.0,(epsilon*(1-ratio))), 1)), float(rb - lb))
rb = lb
if lb == 0:
break
return hatx
def g(*idx):
"""
This function will receive an index tuple from numpy.fromfunction
It's behavior depends on grid_shape: take (i,j) and divide by grid_shape (in each dimension)
That becomes an identifier of the block; then assign a unique integer to it using pairing.
"""
x = numpy.array(idx)
y = numpy.array(grid_shape)
return general_pairing(util.old_div(
x, y)) # broadcasting integer division
def Run(self, W, x, eps, seed):
domain_dimension = len(self.domain_shape)
eps_share = util.old_div(float(eps), domain_dimension)
x = x.flatten()
prng = np.random.RandomState(seed)
Ms = []
ys = []
scale_factors = []
for i in range(domain_dimension):
# Reducde domain to get marginals
marginal_mapping = mapper.MarginalPartition(
domain_shape=self.domain_shape, proj_dim=i).mapping()
reducer = transformation.ReduceByPartition(marginal_mapping)
x_i = reducer.transform(x)
if self.domain_shape[i] < 50:
# run identity subplan
M_i = selection.Identity(x_i.shape).select()
y_i = measurement.Laplace(M_i, eps_share).measure(x_i, prng)
noise_scale_factor = laplace_scale_factor(
M_i, eps_share)
else:
# run dawa subplan
W = get_matrix(W)
W_i = W * support.expansion_matrix(marginal_mapping)
dawa = pmapper.Dawa(eps_share, self.ratio, self.approx)
mapping = dawa.mapping(x_i, prng)
reducer = transformation.ReduceByPartition(mapping)
x_bar = reducer.transform(x_i)
W_bar = W_i * support.expansion_matrix(mapping)
M_bar = selection.GreedyH(x_bar.shape, W_bar).select()
y_i = measurement.Laplace(
M_bar, eps_share * (1 - self.ratio)).measure(x_bar, prng)
noise_scale_factor = laplace_scale_factor(
M_bar, eps_share * (1 - self.ratio))
# expand the dawa reduction
M_i = M_bar * support.reduction_matrix(mapping)
MM = M_i * support.reduction_matrix(marginal_mapping)
Ms.append(MM)
ys.append(y_i)
scale_factors.append(noise_scale_factor)
x_hat = inference.LeastSquares(method='lsmr').infer(Ms, ys, scale_factors)
return x_hat
def statistics(self):
assert self.hist is not None
assert self.edges is not None
hist_data = {}
hist_data['nz_perc'] = util.old_div(np.count_nonzero(self.hist),
float(self.hist.size))
hist_data['max_bin_val'] = self.hist.max()
hist_data['total_records'] = self.hist.sum()
return hist_data
def quad_split_range(cur_range_l, cur_range_u, **kwarg):
'''
Given an rectangular domain represented using boarder cordinates (upper_left, lower_right),
it splits it correctly to 4 quads in the midpoints
'''
ul, lr = cur_range_l, cur_range_u
upper, left = ul
lower, right = lr
n_rows = lower - upper + 1
n_cols = right - left + 1
# If ncol is odd, do vert splits in balanced manner
col_parity = 0
if n_cols % 2:
col_parity = 1
col_midpoint = left + util.old_div(n_cols, 2)
row_midpoint = upper + util.old_div(n_rows, 2)
if n_rows == 1:
# if x has only one row then do only vertical split
row = lr[0]
return [ul, (row, col_midpoint)], [(row, col_midpoint - 1), lr]
if n_cols == 1:
# if x has only one col then do only horizontal split
col = lr[1]
return [ul, (row_midpoint, col)], [(row_midpoint - 1, col), lr]
# o/w do both splits
q1 = (ul, (row_midpoint - 1, col_midpoint - 1))
q2 = ((upper, col_midpoint), (row_midpoint - 1, right))
q3 = ((row_midpoint, left), (lower, col_midpoint - 1 + col_parity))
q4 = ((row_midpoint, col_midpoint + col_parity), lr)
lower = [coordinates[0] for coordinates in [q1, q2, q3, q4]]
upper = [coordinates[1] for coordinates in [q1, q2, q3, q4]]
return lower, upper
def GenerateCells(n, m, num1, num2, grid):
'''
Generate grid shaped celles for UniformGrid and AdaptiveGrid.and
n, m: 2D domain shape
num1, num2: number of cells along two dimensions
grid: grid size
'''
assert math.ceil(util.old_div(n, float(grid))) == num1 and math.ceil(
util.old_div(m, float(grid))
) == num2, "Unable to generate cells for Ugrid: check grid number and grid size"
lower, upper = [], []
for i in range(num1):
for j in range(num2):
lb = [int(i * grid), int(j * grid)]
rb = [int((i + 1) * grid - 1), int((j + 1) * grid - 1)]
if rb[0] >= n:
rb[0] = int(n - 1)
if rb[1] >= m:
rb[1] = int(m - 1)
lower.append(lb)
upper.append(rb)
return lower, upper
def Run(self, W, x, eps):
domain_dimension = len(self.domain_shape)
eps_share = util.old_div(float(eps), domain_dimension)
Ms = []
ys = []
scale_factors = []
for i in range(domain_dimension):
# Reducde domain to get marginals
marginal_mapping = marginal_partition(self.domain_shape, i)
x_i = x.reduce_by_partition(marginal_mapping)
if self.domain_shape[i] < 50:
# run identity subplan
M_i = identity((self.domain_shape[i], ))
y_i = x_i.laplace(M_i, eps_share)
noise_scale_factor = laplace_scale_factor(M_i, eps_share)
else:
# run dawa subplan
W_i = W * support.expansion_matrix(marginal_mapping)
mapping = x_i.dawa(self.ratio, self.approx, eps_share)
x_bar = x_i.reduce_by_partition(mapping)
W_bar = W_i * support.expansion_matrix(mapping)
M_bar = greedyH((len(set(mapping)), ), W_bar)
y_i = x_bar.laplace(M_bar, eps_share * (1 - self.ratio))
noise_scale_factor = laplace_scale_factor(
M_bar, eps_share * (1 - self.ratio))
# expand the dawa reduction
M_i = M_bar * support.reduction_matrix(mapping)
# TODO: Ideally this would be just M_i * support.reduction_matrix(marginal_mapping)
# but currently that returns an int type matrix
# because the type of P_i is int
MM = (support.reduction_matrix(marginal_mapping).T * M_i.T).T
Ms.append(MM)
ys.append(y_i)
scale_factors.append(noise_scale_factor)
x_hat = least_squares(Ms, ys, scale_factors)
return x_hat
def get_boarder(dim_len, branching):
if branching > dim_len:
split_num = dim_len
boarder = [(i, i) for i in range(split_num)]
elif dim_len % branching != 0:
new_hsize = np.divide(float(dim_len), branching)
split_num = [
np.ceil(new_hsize * (i + 1)).astype(int)
for i in range(branching - 1)
]
temp = [i - 1 for i in split_num]
boarder = list(zip(([0] + split_num), (temp + [dim_len - 1])))
else:
cell_size_h = util.old_div(dim_len, branching)
boarder = [(i * cell_size_h, (i + 1) * cell_size_h - 1)
for i in range(branching)]
return boarder
def test_old_div(self):
self.assertEqual(util.old_div(1, 2), 0)
self.assertEqual(util.old_div(2, 2), 1)
self.assertEqual(util.old_div(3, 2), 1)
self.assertEqual(util.old_div(1, 2.0), 0.5)
x = np.array((1.0,))
y = np.array((2.0,), dtype=np.int_)
z = np.array((2.0,), dtype=np.float_)
w = 1.0
zero = np.zeros((1,))
half = 0.5 * np.ones((1,))
self.assertEqual(util.old_div(x, y), zero)
self.assertEqual(util.old_div(x, z), half)
self.assertEqual(util.old_div(w, z), np.array(half))
def Run(self, Q, x, epsilon, seed):
"""Run three engines in order with given epsilons to estimate a
dataset x to answer query set Q
Q - the query workload
x - the underlying dataset
epsilon - the total privacy budget
"""
assert seed is not None, 'seed must be set'
prng = numpy.random.RandomState(seed)
n = len(x)
pSeed = prng.randint(500000)
eSeed = prng.randint(500000)
if self._partition_engine is None:
# ignore ratio when partition_engine is omitted
return self._DirectRun(Q, x, epsilon, eSeed)
else:
if self._ratio < 0 or self._ratio >= 1:
raise ValueError('ratio must in range [0, 1)')
partition = self.Compute_partition(x, epsilon, pSeed)
# check that partition buckets span domain
assert min(itertools.chain(*partition)) == 0
assert max(itertools.chain(*partition)) == (n - 1)
eps2 = (
1 - self._ratio
) * epsilon # this is epsilon_2 used in paper (the epsilon for estimation)
devs = abs(numpy.array(x) - (util.old_div(sum(x), float(len(x)))))
counts = self._estimate_engine.Run(
self._workload_reform(Q, partition, n),
self._dataset_reform(x, partition),
epsilon * (1 - self._ratio), eSeed)
return self._rebuild(partition, counts, n)
def __init__(self, uniformity, dom_shape, scale, seed=None):
'''
Generate synthetic data of varying uniformity
uniformity: parameter in [0,1] where 1 produces perfectly uniform data, 0 is maximally non-uniform
All cells set to zero except fraction equal to 'uniformity' value.
All non-zero cells are set to same value, then shuffled randomly.
'''
self.init_params = util.init_params_from_locals(locals())
self.u = uniformity
assert 0 <= uniformity and uniformity <= 1
n = numpy.prod(dom_shape) # total domain size
hist = numpy.zeros(n)
num_nonzero = max(1, int(uniformity * n))
hist_value = util.old_div(scale, num_nonzero)
hist[0:num_nonzero] = hist_value
prng = numpy.random.RandomState(seed)
prng.shuffle(hist)
super(DatasetUniformityVaryingSynthetic,
self).__init__(hist.reshape(dom_shape),
reduce_to_domain_shape=None,
dist=None)
def infer(self, Ms, ys, scale_factors=None):
''' Either:
1) Ms is a single M and ys is a single y
(scale_factors ignored) or
2) Ms and ys are lists of M matrices and y vectors
and scale_factors is a list of the same length.
'''
A, y = self._apply_scales(Ms, ys, scale_factors)
if self.known_total is not None:
A, y = self.__known_total_problem(A, y)
if self.method == 'standard':
assert self.l2_reg == 0, 'l2 reg not supported with method=standard'
assert isinstance(
A,
np.ndarray), "method 'standard' only works with dense matrices"
(x_est, _, rank, _) = linalg.lstsq(A, y, lapack_driver='gelsy')
elif self.method == 'lsmr':
res = lsmr(A, y, atol=0, btol=0, damp=self.l2_reg)
x_est = res[0]
elif self.method == 'lsqr':
res = lsqr(A, y, atol=0, btol=0, damp=self.l2_reg)
x_est = res[0]
if self.known_total is not None:
x_est = np.append(x_est, self.known_total - x_est.sum())
x_est = x_est.reshape(A.shape[1]) # reshape to match shape of x
# James-Stein estimation
if self.stein and x_est.size >= 3:
adjustment = 1.0 - util.old_div((x_est.size - 2), (x_est**2).sum())
x_est *= adjustment
return x_est
def variance(N, b):
'''Computes variance given domain of size N
and branchng factor b. Equation 3 from paper.'''
h = math.ceil(math.log(N, b))
return (((b - 1) * h**3) - (util.old_div((2 * (b + 1) * h**2), 3)))
def cantor_pairing(a, b):
"""
A function returning a unique positive integer for every pair (a,b) of positive integers
"""
return util.old_div((a + b) * (a + b + 1), 2) + b
def fractionZeros(self):
zero_count = (self.payload == 0).sum()
return util.old_div(float(zero_count), self.payload.size)
It's behavior depends on grid_shape: take (i,j) and divide by grid_shape (in each dimension)
That becomes an identifier of the block; then assign a unique integer to it using pairing.
"""
x = numpy.array(idx)
y = numpy.array(grid_shape)
return general_pairing(util.old_div(
x, y)) # broadcasting integer division
h = numpy.vectorize(g)
# numpy.fromfunction builds an array of domain_shape by calling a function with each index tuple (e.g. (i,j))
partition_array = numpy.fromfunction(h, domain_shape, dtype=int)
# transform to canonical order
partition_array = canonicalTransform(partition_array)
return partition_array
if __name__ == '__main__':
scale = 10000
for u in [util.old_div(i, 10.0) for i in range(0, 11)]:
print(u)
d = DatasetUniformityVaryingSynthetic(uniformity=u,
dom_shape=(10, ),
scale=scale,
seed=999)
size = d.payload.size
unif = numpy.empty_like(d.payload)
unif.fill(util.old_div(scale, float(size)))
print(sum(abs(unif - d.payload)))
def _GreedyHierByLv(self, fullQtQ, n, offset, depth=0, withRoot=False):
"""Compute the weight distribution of one node of the tree by minimzing
error locally.
fullQtQ - the same matrix as QtQ in the Run method
n - the size of the submatrix that is corresponding
to current node
offset - the location of the submatrix in fullQtQ that
is corresponding to current node
depth - the depth of current node in the tree
withRoot - whether the accurate root count is given
Returns: error, inv, weights, queries
error - the variance of query on current node with epsilon=1
inv - for the query strategy (the actrual weighted queries to be asked)
matrix A, inv is the inverse matrix of A^TA
weights - the weights of queries to be asked
queries - the list of queries to be asked (all with weight 1)
"""
if n == 1:
return numpy.linalg.norm(fullQtQ[:, offset], 2)**2, \
numpy.array([[1.0]]), \
numpy.array([1.0]), [[offset, offset]]
QtQ = fullQtQ[:, offset:offset + n]
if (numpy.min(QtQ, axis=1) == numpy.max(QtQ, axis=1)).all():
mat = numpy.zeros([n, n])
mat.fill(util.old_div(1.0, n**2))
return numpy.linalg.norm(QtQ[:,0], 2)**2, \
mat, numpy.array([1.0]), [[offset, offset+n-1]]
if n <= self._branch:
bound = list(zip(list(range(n)), list(range(1, n + 1))))
else:
rem = n % self._branch
step = util.old_div((n - rem), self._branch)
swi = (self._branch - rem) * step
sep = list(range(0, swi, step)) + list(range(swi, n,
step + 1)) + [n]
bound = list(zip(sep[:-1], sep[1:]))
serr, sinv, sdist, sq = list(
zip(*[
self._GreedyHierByLv(
fullQtQ, c[1] - c[0], offset + c[0], depth=depth + 1)
for c in bound
]))
invAuList = [c.sum(axis=0) for c in sinv]
invAu = numpy.hstack(invAuList)
k = invAu.sum()
m1 = sum(
map(
lambda rng, v: numpy.linalg.norm(
numpy.dot(QtQ[:, rng[0]:rng[1]], v), 2)**2, bound,
invAuList))
m = numpy.linalg.norm(numpy.dot(QtQ, invAu), 2)**2
sumerr = sum(serr)
if withRoot:
return sumerr, block_diag(*sinv), \
numpy.hstack([[0], numpy.hstack(sdist)]), \
[[offset, offset+n-1]] + list(itertools.chain(*sq))
decay = util.old_div(1.0, (self._branch**(util.old_div(depth, 2.0))))
err1 = numpy.array(list(range(self._granu, 0, -1)))**2
err2 = numpy.array(list(range(self._granu)))**2 * decay
toterr = 1.0 / err1 * (sumerr - ((m - m1) * decay + m1) * err2 /
(err1 + err2 * k))
err = toterr.min() * self._granu**2
perc = 1 - util.old_div(numpy.argmin(toterr), float(self._granu))
inv = (util.old_div(1.0, perc))**2 * (
block_diag(*sinv) - (1 - perc)**2 / (perc**2 + k * (1 - perc)**2) *
numpy.dot(invAu.reshape([n, 1]), invAu.reshape([1, n])))
dist = numpy.hstack([[1 - perc], perc * numpy.hstack(sdist)])
return err, inv, dist, \
[[offset, offset+n-1]] + list(itertools.chain(*sq))
def testDatasetReduce(self):
div = 4
new_shape = (util.old_div(self.hist.shape[0], div), )
dr = dataset.Dataset(hist=self.hist, reduce_to_domain_shape=new_shape)
self.assertEqual(dr.domain_shape, new_shape)
