def _contract(self, ket_1, ket_2, gate, direction):
# Contract the tensor
tens = np.tensordot(np.tensordot(ket1, gate, axes=[0, 2]),
ket2,
axes=([1, 4], [1, 0]))
tens = tens.reshape((tens.shape[0] * tens.shape[1], -1))
# Perform the SVD
u, s, v = np.linalg.svd(tens, full_matrices=False)
# Restrict to the pre-determined bond dimension ``self.bond_dim``
u = u[:, :self.bond_dim]
s = s[:self.bond_dim]
v = v[:self.bond_dim, :]
# Combine ``s`` with either ``u`` or ``v`` depending on whether direction is left or right, respectively
if direction == "right":
v = np.dot(np.diag(s), v)
elif direction == "left":
u = np.dot(u, np.diag(s))
else:
raise ValueError(
"Direction must be one of ['right', 'left'], but has value {}".
format(direction))
# Keep coefficient of unity
u = u / np.amax(u)
v = v / np.amax(v)
ket_1 = np.tranpose(u.reshape((-1, d, u.shape[1])), (1, 0, 2))
ket_2 = np.tranpose(v.reshape((v.shape[1], d, -1)), (1, 0, 2))
return ket_1, ket_2
def corrnoise(mn=None, R=_default_R, n=100):
if mn is None:
mn = np.zeros(np.shape(R[0])[0])
nvect = np.random.randn(np.shape(R)[0], n)
R = 0.5 * (R + np.tranpose(R))
v, d = np.linalg.eig(R)
if np.any(np.diag <= 0):
print('R must be positive definite')
w = v * np.sqrt(d)
nc = w * nvect
Rout = (nc * np.tranpose(nc)) / n
return nc, Rout
def compute_accels_vectors(self, vels_vectors, time_checkpoints):
x_vels, y_vels, z_vels = np.tranpose(vels_vectors)
x_accels = self.compute_derivative(time_checkpoints, vels_vectors)
y_accels = self.compute_derivative(time_checkpoints, vels_vectors)
z_accels = self.compute_derivative(time_checkpoints, vels_vectors)
accels_vectors = np.transpose([x_accels, y_accels, z_accels])
return accels_vectors
def feature_sign(A, y, g):
y = np.array(y)
dim = np.shape(A)
x = np.zeros([dim(1), 1])
theta = np.sign(x)
act_set = []
for i in range(100):
D = 0
index = 0
for j in range(dim(1)):
ai = np.tranpose(A)[j]
ai = np.reshape(ai, [np.shape(ai)[0], 1])
d = np.matmul(ai, y)
d = d - (np.matmul(ai, np.matmul(A, y)))
d = abs(d)
if D < d:
index = j
D = d
act_set.append(index)
if D > g:
theta[index] = -1
else:
theta[index] = 1
act_set.append(i)
return x
def collectDaymet(tmpdir, bounds, start, end, vars=None, force=False):
"""Calls the DayMet Rest API to get data and save raw data."""
if vars is None:
vars = VALID_VARIABLES
dat = dict()
d_inited = False
for year in range(start.year, end.year + 1):
for var in vars:
fname = downloadFile(tmpdir, bounds, year, var, force)
x, y, v = loadFile(fname, var)
if not d_inited:
initData(dat, vars, numDays(start, end), len(x), len(y))
d_inited = True
# stuff v in the right spot
if year == start.year and year == end.year:
dat[var][:, :, :] = np.transpose(
v[start.doy:end.doy + 1, :, :], (0, 2, 1))
elif year == start.year:
dat[var][0:365 - start.doy + 1, :, :] = np.transpose(
v[start.doy - 1:, :, :], (0, 2, 1))
elif year == end.year:
dat[var][-end.doy:, :, :] = np.transpose(
v[-end.doy:, :, :], (0, 2, 1))
else:
my_start = 365 * (year - start.year) - start.doy + 1
dat[var][my_start:my_start + 365, :, :] = np.tranpose(
v, (0, 2, 1))
return dat
def DiffOper(N):
D = sp.diags([-np.ones(N, 1), np.ones(N, 1)], [0, 1], N, N + 1)
D[:, 1] = []
D[(1, 1)] = 0
B = [[np.kron(sp.eye(N), D)], [np.kron(D, sp.eye(N))]]
Bt = np.tranpose(B)
BtB = Bt * B
return (B, Bt, BtB)
def definit_R_informal(Uo, param, trans_tmp_cout_generalise, income, amenity):
R_mat = 1 / param["size_shack"] * (
income - trans_tmp.cout_generalise -
(repmat(np.tranpose(Uo), 1, np.size(income, 2)) /
(amenity *
(param["size_shack"] - param["q0"])**param["coeff_beta"]))**
(1 / param["coeff_alpha"]))
return R_mat
def omeFromMRP(mrp, mrpRates):
sqNor = np.linalg.norm(mrp)**2
matMRP = np.array([[
1 - sqNor + 2 * mrp[0]**2, 2 * (mrp[0] * mrp[1] - mrp[2]),
2 * (mrp[0] * mrp[2] - mrp[1])
],
[
2 * (mrp[0] * mrp[1] + mrp[2]),
1 - sqNor + 2 * mrp[1]**2,
2 * (mrp[2] * mrp[1] - mrp[0])
],
[
2 * (mrp[0] * mrp[2] + mrp[1]),
2 * (mrp[2] * mrp[1] + mrp[0]),
1 - sqNor + 2 * mrp[2]**2
]])
return np.dot(np.tranpose(matMRP), mrpRates) * 4 / (1 + sqNor)**2
def attack(args):
outputs = construct_model_and_data(args)
data_model = outputs['data_model']
x_test = outputs['x_test']
y_test = outputs['y_test']
model = outputs['model']
clip_max = outputs['clip_max']
clip_min = outputs['clip_min']
if args.attack_type == 'targeted':
target_labels = outputs['target_labels']
target_images = outputs['target_images']
for i, sample in enumerate(x_test[:args.num_samples]):
label = y_test[i] #np.argmax(y_test[i])
if args.attack_type == 'targeted':
target_label = target_labels[i]
target_image = target_images[i]
else:
target_label = None
target_image = outputs['target_images'][i]
print('attacking the {}th sample...'.format(i))
perturbed = hsja(model,
sample,
clip_max=1,
clip_min=0,
constraint=args.constraint,
num_iterations=args.num_iterations,
gamma=1.0,
target_label=target_label,
target_image=target_image,
stepsize_search=args.stepsize_search,
max_num_evals=1e4,
init_num_evals=100)
if np.argmin(sample.shape) == 0:
sample = np.transpose(sample, (1, 2, 0))
if np.argmin(perturbed.shape) == 0:
perturbed = np.tranpose(perturbed, (1, 2, 0))
image = np.concatenate(
[sample, np.zeros((32, 8, 3)), perturbed], axis=1)
imageio.imsave(
'{}/figs/{}-{}-{}.jpg'.format(data_model, args.attack_type,
args.constraint, i), image)
def reshape_binned(x, bin_counts, reflect=False):
"""Reshape an array of `(n_samples, prod(n_bins_dim))` into `(n_samples, n_bins_dim0, n_bins_dim1, ...)`
"""
x = x.reshape(-1, *bin_counts)
ndims = len(bin_counts)
if reflect:
if ndims == 1: # Flip x
x = np.flip(x, axis=1)
if ndims == 2: # Swap x and y
x = np.transpose(x, (0, 2, 1))
else:
if isinstance(reflect, tuple) and len(reflect == ndims + 1):
x = np.tranpose(x, reflect)
else:
raise ValueError(
'For a n > 2 dimensional reflection, reflect must be a tuple for np.tranpose'
)
return x
def sudoku_checker(board):
# checking numbers in each row
for row in board:
dup_check = set(row)
if len(dup_check) != len(row):
return False
# checking numbers in each column
tranposed_board = numpy.tranpose(board)
for col in tranposed_board:
dup_check = set(col)
if len(dup_check) != len(col):
return False
# checking in 3x3 quadrant
for row in (0, 3, 6):
for col in (0, 3, 6):
quadrant = [board[x][y] for x in range(row, row + 3) for y in range(col, col + 3)]
dup_check = set(quadrant)
if len(dup_check) != len(quadrant):
return False
return True
def read(ifile, **kwds):
# ifile is a list, generate Feature object by iterating through list
# Pixelsize must be function input (not kwd) by the time we call 'Feature'.
arr = np.array([dicom.read_file(frame).pixel_array for frame in ifile])
arr = np.tranpose(arr, axes=(1, 2, 0))
if 'pixelsize' in kwds:
px_size = float(kwds['pixelsize'])
else:
# Ensure that all frames share similar pixel sizes
sizes = np.array([dicom.read_file(f).PixelSpacing for f in ifile],
dtype=float)
sizes = np.ravel(sizes)
if not np.allclose(sizes, sizes[0]):
msg = 'Voxel dimensions are not the same.'
raise NotImplementedError(msg)
px_size = 1000 * sizes[0]
warn('Pixelsize for sample {} has been set to: {}'.format(
ifile[0], px_size))
return Feature(arr, pixelsize=px_size)
def J(xi, w_bar, N, Q_C, t):
# create error vector
# fill up error vector with all training data
e = np.empty([N,1])
for i in range(N):
e.append(error_i(xi[i], xi[i+1],w_bar[i],w_bar[i+1]))
# create Q matrix and invert it
# create a vector of all Q_i entries
Q_vect = np.empty([N,1])
for i in range(N):
Q_vect.append(Q_i(Q_C(6),t[i],t[i-1]))
Q = np.diag(Q_vect)
Q_inv = np.linalg.inv(Q)
return (1/2)*np.matmul(np.matmul(np.tranpose(e),Q_inv),e)
def computeLayer(X, W, b):
S = np.dot(np.tranpose(W),X) + b
return S
def transpose(a, axes=None):
try:
return a.transpose(axes)
except AttributeError:
return numpy.tranpose(a, axes)
def setPoints(self, points): self.points = points del self.area x, y = np.tranpose(points) self.setData(x=x, y=y, symbolSize=2, symbol='o', symbolPen=(0, 0, 255), pen=None) self.setBorderVisibility(True)
