def TrackCurvature(x):
# use quadratic b-splines at each point to estimate curvature
# i get almost the same formula i had before but it's off by a factor of 4!
xx = np.concatenate([x[-1:], x, x[:1]])
p0 = xx[:-2]
p1 = xx[1:-1]
p2 = xx[2:]
T = p2 - p0 # track derivative
uT = np.abs(T)
TT = 4*(p0 - 2*p1 + p2)
k = (np.real(T)*np.imag(TT) - np.imag(T)*np.real(TT)) / (uT**3)
return k
def test_unimplemented_falseyness():
@contextmanager
def remove_grad_definitions(fun):
vjpmaker = primitive_vjps.pop(fun, None)
yield
if vjpmaker:
primitive_vjps[fun] = vjpmaker
with remove_grad_definitions(np.iscomplex):
fun = lambda x: np.real(x**2 if np.iscomplex(x) else np.sum(x))
check_grads(fun)(5.)
check_grads(fun)(2. + 1j)
def fit_gaussian_draw(X, J, seed=28, reg=1e-7, eig_pow=1.0):
"""
Fit a multivariate normal to the data X (n x d) and draw J points
from the fit.
- reg: regularizer to use with the covariance matrix
- eig_pow: raise eigenvalues of the covariance matrix to this power to construct
a new covariance matrix before drawing samples. Useful to shrink the spread
of the variance.
"""
with NumpySeedContext(seed=seed):
d = X.shape[1]
mean_x = np.mean(X, 0)
cov_x = np.cov(X.T)
if d==1:
cov_x = np.array([[cov_x]])
[evals, evecs] = np.linalg.eig(cov_x)
evals = np.maximum(0, np.real(evals))
assert np.all(np.isfinite(evals))
evecs = np.real(evecs)
shrunk_cov = evecs.dot(np.diag(evals**eig_pow)).dot(evecs.T) + reg*np.eye(d)
V = np.random.multivariate_normal(mean_x, shrunk_cov, J)
return V
def as_scalar(x):
vs = vspace(getval(x))
if vs.iscomplex:
x = np.real(x)
if vs.shape == ():
return x
elif vs.size == 1:
return x.reshape(())
else:
raise TypeError(
"Output {} can't be cast to float. "
"Function grad requires a scalar-valued function. "
"Try jacobian or elementwise_grad.".format(getval(x)))
def QuadFitCurvatureMap(x):
curv = []
for i in range(len(x)):
# do a look-ahead quadratic fit, just like the car would do
pts = x[(np.arange(6) + i) % len(x)] / 100 # convert to meters
basis = (pts[1] - pts[0]) / np.abs(pts[1] - pts[0])
# project onto forward direction
pts = (np.conj(basis) * (pts - pts[0]))
p = np.polyfit(np.real(pts), np.imag(pts), 2)
curv.append(p[0] / 2)
return np.float32(curv)
def log_likelihood(Data,
frequencies,
DL,
t_c,
phi_c,
chirpm,
symmratio,
spin1,
spin2,
alpha_squared,
bppe,
NSflag,
N_detectors,
detector,
cosmology=cosmology.Planck15):
#Z = Distance(DL/mpc,unit=u.Mpc).compute_z(cosmology = cosmology)
#chirpme = chirpm/(1+Z)
mass1 = utilities.calculate_mass1(chirpm, symmratio)
mass2 = utilities.calculate_mass2(chirpm, symmratio)
model = dcsimr_detector_frame(mass1=mass1,
mass2=mass2,
spin1=spin1,
spin2=spin2,
collision_time=t_c,
collision_phase=phi_c,
Luminosity_Distance=DL,
phase_mod=alpha_squared,
cosmo_model=cosmology,
NSflag=NSflag,
N_detectors=N_detectors)
frequencies = np.asarray(frequencies)
amp, phase, hreal = model.calculate_waveform_vector(frequencies)
#h_complex = np.multiply(amp,np.add(np.cos(phase),-1j*np.sin(phase)))
h_complex = amp * np.exp(-1j * phase)
noise_temp, noise_func, freq = model.populate_noise(detector=detector,
int_scheme='quad')
resid = np.subtract(Data, h_complex)
#integrand_numerator = np.multiply(np.conjugate(Data), h_complex) + np.multiply(Data,np.conjugate( h_complex))
integrand_numerator = np.multiply(resid, np.conjugate(resid))
noise_root = noise_func(frequencies)
noise = np.multiply(noise_root, noise_root)
integrand = np.divide(integrand_numerator, noise)
integral = np.real(simps(integrand, frequencies))
return -2 * integral
def grad_k(ws, fdensity, alpha, sig, psf_k):
#print('grad_k begin');
mo = np.exp(-4.)
ws = real_to_complex(ws)
ws = ws.reshape((n_grid, n_grid))
#wk = ws;
#ws = np.real(fft.ifft2(ws));
l1 = -1 * fft.ifft2(
(np.real(fft.ifft2(ws * psf_k)) - data) / sig_noise**2) * psf_k
#print(l1-l1_other)
l1 = l1.flatten()
l_tot = l1
#return l1,l2;
l_tot = complex_to_real(l_tot)
#print('grad is');
#print(l_tot);
return l_tot
def list_simulate_spectral(cov, J, n_simulate=1000, seed=82):
"""
Simulate the null distribution using the spectrums of the covariance
matrix. This is intended to be used to approximate the null
distribution.
Return (a numpy array of simulated n*FSSD values, eigenvalues of cov)
"""
# eigen decompose
eigs, _ = np.linalg.eig(cov)
eigs = np.real(eigs)
# sort in decreasing order
eigs = -np.sort(-eigs)
sim_fssds = FSSD.simulate_null_dist(eigs,
J,
n_simulate=n_simulate,
seed=seed)
return sim_fssds, eigs
def best_invec_phase(x, y, **kwargs):
"""Computes the l2-distance between `x` and `y` up to a global phasefactor.
.. math::
\min_\phi \Vert x - \mathrm{e}^{i \phi} y \Vert_2
:param x, y: Input vectors of same length
:param kwargs: Parameters passed to `scipy.optimize.minimize`
:returns: Minimal distane (and possibly optimal vector `y`)
"""
norm_sq = lambda x: np.real(np.dot(np.conj(x.ravel()), x.ravel()))
cost = lambda phi: norm_sq(x - np.exp(1.j * phi) * y)
# Choose initialization randomly to evade maximimum at opposite side
result = minimize(cost, rand_angles(), jac=grad(cost), **kwargs)
y_ = np.exp(1.j * result['x']) * y
return y_, result['fun']
def forward(self, x):
x = torch.tensor(npa.real(npa.fft.fft(x.to('cpu').numpy(),
axis=2))).to('cuda')
x, edge_index = x, self.coos[0].to(x.device)
x = self.conv1(x, edge_index)
x = F.relu(x)
x = gcn_pool_4(x)
edge_index = self.coos[2].to(x.device)
x = self.conv2(x, edge_index)
x = F.relu(x)
x = gcn_pool_4(x)
x = x.view(x.shape[0], -1)
x = self.fc1(x)
x = F.relu(x)
x = self.fc2(x)
return F.log_softmax(x, dim=1)
def nn_predict_tgcn_cheb(params, x):
L = graph.rescale_L(hyper['L'][0], lmax=2)
w = np.fft.fft(x, axis=2)
xc = chebyshev_time_vertex(L, w, hyper['filter_order'])
y = np.einsum('knhq,kfh->fnq', xc, params['W1'])
y += np.expand_dims(params['b1'], axis=2)
# nonlinear layer
# y = np.tanh(y)
y = ReLU(y)
# dense layer
y = np.einsum('fnq,cfn->cq', y, params['W2'])
y += np.expand_dims(params['b2'], axis=1)
outputs = np.real(y.T)
return outputs - logsumexp(outputs, axis=1, keepdims=True)
def grad_like(self, wsp, ws, ws_k, xi):
#print('start grad_like')
conv = np.real(fft.ifft2(ws_k * self.psf_k))
#convolution of ws with psf
term1 = (
conv - self.data
) / self.n_grid**2 / self.sig_noise**2 #term thats squared in like (with N in denom)
grad = np.zeros((self.n_grid, self.n_grid), dtype='complex')
for i in range(0, self.n_grid):
for j in range(0, self.n_grid):
#try to modulate by hand
ft1 = fft.fft2(1 / (1 + np.exp(-1 * wsp / xi)))
ftp = np.roll(ft1, (i, j), axis=(0, 1))
term2 = fft.ifft2(ftp * self.psf_k)
grad[i, j] = np.sum(term1 * term2)
grad_real = self.complex_to_real(np.conj(grad.flatten()))
#embed to 2R
#print('end grad_like');
return grad_real
def rcwa_assembly(dofold,freq,theta,phi,planewave):
'''
planewave:{'p_amp',...}
'''
df = f_symmetry(dofold,Mx,My,xsym,ysym,Nlayer=Nlayer)
dof = b_filter(df,bproj)
obj = rcwa.RCWA_obj(nG,L1,L2,freq,theta,phi,verbose=0)
obj.Add_LayerUniform(thick0,epsuniform)
epsdiff=[]
for i in range(Nlayer):
epsdiff.append(mstruct[i].epsilon(lam0/np.real(freq),x_type = 'lambda')-epsbkg)
obj.Add_LayerGrid(thick[i],epsdiff[i],epsbkg,Nx,Ny)
obj.Add_LayerUniform(thickN,epsuniform)
obj.Init_Setup(Gmethod=0)
obj.MakeExcitationPlanewave(planewave['p_amp'],planewave['p_phase'],planewave['s_amp'],planewave['s_phase'],order = 0)
obj.GridLayer_getDOF(dof)
return obj,dof,epsdiff
def cost(self, controls, states, system_eval_step):
"""
Compute the penalty.
Arguments:
controls
states
system_eval_step
Returns:
cost
"""
# The cost is the infidelity of each evolved state and its target state.
inner_products = anp.matmul(self.target_states_dagger, states)[:, 0, 0]
fidelities = anp.real(inner_products * anp.conjugate(inner_products))
fidelity_normalized = anp.sum(fidelities) / self.state_count
infidelity = 1 - fidelity_normalized
return infidelity * self.cost_multiplier
def my_spectral_clustering(sim_mat, n_clusters=2):
N = sim_mat.shape[0]
sim_mat = sim_mat - np.diag(np.diag(sim_mat))
t1 = 1. / np.sqrt(np.sum(sim_mat, axis=1))
t2 = np.dot(t1.reshape(N, 1), t1.reshape(1, N))
lap_mat = np.eye(N) - sim_mat * t2
eig_val, eig_vec = np.linalg.eig(lap_mat)
idx = eig_val.argsort()
eig_val = eig_val[idx]
eig_vec = np.real(eig_vec[:, idx])
t3 = np.diag(np.sqrt(1. / np.sum(eig_vec[:, 0:n_clusters]**2, axis=1)))
embd = np.dot(t3, eig_vec[:, 0:n_clusters])
clf = KMeans(n_clusters=n_clusters, n_jobs=-1)
label_pred = clf.fit_predict(embd)
return label_pred
def my_spectral_clustering(sim_mat, n_clusters=2):
N = sim_mat.shape[0]
sim_mat = sim_mat - np.diag(np.diag(sim_mat))
t1 = 1./np.sqrt(np.sum(sim_mat, axis=1))
t2 = np.dot(t1.reshape(N,1), t1.reshape(1,N))
lap_mat = np.eye(N) - sim_mat * t2
eig_val, eig_vec = np.linalg.eig(lap_mat)
idx = eig_val.argsort()
eig_val = eig_val[idx]
eig_vec = np.real(eig_vec[:, idx])
t3 = np.diag(np.sqrt(1./np.sum(eig_vec[:,0:n_clusters]**2, axis=1)))
embd = np.dot(t3, eig_vec[:, 0:n_clusters])
clf = KMeans(n_clusters = n_clusters, n_jobs=-1)
label_pred = clf.fit_predict(embd)
return label_pred
def pu2r(*tlist):
"""
Returns real part of u2, and registers it as being primitive.
Primitive means that its derivative will be provided in a defvjp (
def of vector-jacobian-product) so no need for autograd to calculate it
from the u2 definition.
Parameters
----------
tlist : list[float]
len = 4
Returns
-------
np.ndarray
shape=(2,2)
"""
return np.real(u2_alt(*tlist))
def forward(self, x):
# X - [n_ex, length, n_in]
# Y - [n_ex, length, n_out]
#Y=np.zeros((X.shape[0],X.shape[1],self.n_out))
lf = self.fir_length
lx = x.shape[1]
x2 = np.concatenate((x, np.zeros((x.shape[0], lf, x.shape[2]))),
axis=1)
f2 = np.concatenate((self.W, np.zeros((lx, self.n_inp, self.n_out))),
axis=0)
X = np.fft.fft(x2, axis=1)
F = np.fft.fft(f2, axis=0)
tmp = np.sum(F[np.newaxis, ...] * X[..., np.newaxis], axis=2)
Y = np.real(np.fft.ifft(tmp, axis=1))[:, 0:x.shape[1]]
Y += self.b.reshape(1, 1, -1)
return Y
def loss(K_conj):
"""
K is a tensor of CONJUGATE Kraus Operators of dim s x w x n x n
s: output_dim
w: ops_per_output
n: state_dim
"""
total_loss = 0.0
# Iterate over each sequence in batch
for i in range(batch.shape[0]):
seq = batch[i]
rho_new = rho.copy()
# burn in
for b in range(burn_in):
temp_rho = np.zeros(
[K_conj.shape[1], K_conj.shape[2], K_conj.shape[3]],
dtype='complex128')
for w in range(K_conj.shape[1]):
temp_rho[w, :, :] = np.dot(
np.dot(K[int(seq[b]) - 1, w, :, :], rho_new),
np.conjugate(K[int(seq[b]) - 1, w, :, :]).T)
rho_new = np.sum(temp_rho, 0)
rho_new = rho_new / np.trace(rho_new)
# Compute likelihood for the sequence
for s in seq[burn_in:]:
rho_sum = np.zeros([K_conj.shape[2], K_conj.shape[2]],
dtype='complex128')
for w in range(K.shape[1]):
# subtract 1 to adjust for MATLAB indexing
rho_sum += np.dot(
np.dot(np.conjugate(K_conj[int(s) - 1, w, :, :]),
rho_new), K_conj[int(s) - 1, w, :, :].T)
rho_new = rho_sum
total_loss += np.log(np.real(np.trace(rho_new)))
return -total_loss / batch.shape[0]
def complex_as_matrix(z, n):
"""Represent a complex number as a matrix.
Parameters
----------
z : complex float
n : int (even)
Returns
-------
Z : ndarray (n,n)
Real-valued n*n tri-diagonal matrix representing z in the ring of n*n matrices.
"""
Z = np.zeros((n, n))
ld = np.zeros(n - 1)
ld[0::2] = np.imag(z)
np.fill_diagonal(Z[1:], ld)
Z = Z - Z.T
np.fill_diagonal(Z, np.real(z))
return Z
def ifft2_and_shift(var_real,
var_imag,
axes=(-2, -1),
override_backend=None,
normalize=False):
bn = override_backend if override_backend is not None else global_settings.backend
if bn == 'autograd':
var = var_real + 1j * var_imag
norm = None if not normalize else 'ortho'
var = anp.fft.fftshift(anp.fft.ifft2(var, axes=axes, norm=norm),
axes=axes)
return anp.real(var), anp.imag(var)
elif bn == 'pytorch':
var = tc.stack([var_real, var_imag], dim=-1)
var = tc.ifft(var, signal_ndim=2, normalized=normalize)
var_real, var_imag = tc.split(var, 1, dim=-1)
slicer = [slice(None)] * (var_real.ndim - 1) + [0]
var_real = var_real[tuple(slicer)]
var_imag = var_imag[tuple(slicer)]
var_real = fftshift(var_real, axes=axes)
var_imag = fftshift(var_imag, axes=axes)
return var_real, var_imag
def simple_2d_filter(x, h):
"""A simple 2d filter algorithm that is differentiable with autograd.
Uses a 2D fft approach since it is typically faster and preserves the shape
of the input and output arrays.
The ffts pad the operation to prevent any circular convolution garbage.
Parameters
----------
x : array_like (2D)
Input array to be filtered. Must be 2D.
h : array_like (2D)
Filter kernel (before the DFT). Must be same size as `x`
Returns
-------
array_like (2D)
The output of the 2d convolution.
"""
(kx, ky) = x.shape
x = _edge_pad(x,((kx, kx), (ky, ky)))
return _centered(npa.real(npa.fft.ifft2(npa.fft.fft2(x)*npa.fft.fft2(h))),(kx, ky))
def best_tmat_phases(A, B, cols=True, rows=True, **kwargs):
"""Finds the angles `phi` and `psi` that minimize the Frobenius distance
between A and B', where
.. math::
B' = \mathrm{diag}(\mathrm{exp}(i \phi)) B \mathrm{diag}(\mathrm{exp}(i \psi))
:returns: Optimal value `B'` as well as minimal Frobenius distance
"""
d = len(A)
diagp = lambda phi: np.diag(np.exp(1.j * phi))
B_ = lambda phi, psi: np.dot(diagp(phi), np.dot(B, diagp(psi)))
norm_sq = lambda x: np.real(np.dot(np.conj(x.ravel()), x.ravel()))
if cols and rows:
cost = lambda x: norm_sq(A - B_(x[:d], x[d:]))
init_angles = rand_angles(2 * d)
elif rows:
cost = lambda x: norm_sq(A - B_(x, np.zeros(d)))
init_angles = rand_angles(d)
elif cols:
cost = lambda x: norm_sq(A - B_(np.zeros(d), x))
init_angles = rand_angles(d)
else:
raise ValueError('Either rows or cols should be true')
result = minimize(cost, init_angles, jac=grad(cost), **kwargs)
if cols and rows:
phi, psi = result['x'][:d], result['x'][d:]
elif rows:
phi, psi = result['x'], np.zeros(d)
elif cols:
phi, psi = np.zeros(d), result['x']
# Normalization 1 / np.sqrt(2) due to real embedding
return B_(phi, psi), result['fun'] / np.sqrt(2)
def match(self, scene):
# 1) determine shape of scene in obs, set mask
# 2) compute the interpolation kernel between scene and obs
# 3) compute obs.psf in the frame of scene, store in Fourier space
# A few notes on this procedure:
# a) This assumes that scene.psfs and self.psfs have the same spatial shape,
# which will need to be modified for multi-resolution datasets
if self._psfs is not None:
ipad, ppad = interpolation.get_common_padding(self.images,
self._psfs,
padding=self.padding)
self.image_padding, self.psf_padding = ipad, ppad
_psfs = np.pad(self._psfs, ((0, 0), *self.psf_padding), 'constant')
_target = np.pad(scene._psfs, self.psf_padding, 'constant')
new_kernel_fft = []
# Deconvolve the target PSF
target_fft = np.fft.fft2(np.fft.ifftshift(_target))
for _psf in _psfs:
observed_fft = np.fft.fft2(np.fft.ifftshift(_psf))
# Create the matching kernel
kernel_fft = observed_fft / target_fft
# Take the inverse Fourier transform to normalize the result
# Trials without this operation are slow to converge, but in the future
# we may be able to come up with a method to normalize in the Fourier Transform
# and avoid this step.
kernel = np.fft.ifft2(kernel_fft)
kernel = np.fft.fftshift(np.real(kernel))
kernel /= kernel.sum()
# Store the Fourier transform of the matching kernel
new_kernel_fft.append(np.fft.fft2(np.fft.ifftshift(kernel)))
self.psfs_fft = np.array(new_kernel_fft)
return self
def g_out_antihess(y):
lp = snp_log_probs(y)
ret = 0.0
for l in seg_sites._get_likelihood_sequences(lp):
L = len(l)
lc = make_constant(l)
fft = np.fft.fft(l)
# (assumes l is REAL)
assert np.all(np.imag(l) == 0.0)
fft_rev = np.conj(fft) * np.exp(
2 * np.pi * 1j * np.arange(L) / float(L))
curr = 0.5 * (fft * fft_rev - fft * make_constant(fft_rev) -
make_constant(fft) * fft_rev)
curr = np.fft.ifft(curr)[(L - 1)::-1]
# make real
assert np.allclose(np.imag(curr / L), 0.0)
curr = np.real(curr)
curr = curr[0] + 2.0 * np.sum(curr[1:int(np.sqrt(L))])
ret = ret + curr
return ret
def fft_convolve(*images):
"""Use FFT's to convove an image with a kernel
Parameters
----------
images: list of array-like
A list of images to convolve.
Returns
-------
result: array
The convolution in pixel space of `img` with `kernel`.
"""
from autograd.numpy.numpy_boxes import ArrayBox
Images = [np.fft.fft2(np.fft.ifftshift(img)) for img in images]
if np.any([isinstance(img, ArrayBox) for img in images]):
Convolved = Images[0]
for img in Images[1:]:
Convolved = Convolved * img
else:
Convolved = np.prod(Images, 0)
convolved = np.fft.ifft2(Convolved)
return np.fft.fftshift(np.real(convolved))
def realify(Y):
"""Convert data in k-dimensional complex space to 2k-dimensional
real space.
Parameters
----------
Y : ndarray (k,n)
Real-valued array of data, `k` must be even.
Returns
-------
Yreal : ndarray (2k,n)
Complex-valued array of data.
"""
if Y.ndim == 1:
Yreal = np.zeros(2 * Y.shape[0])
else:
Yreal = np.zeros((2 * Y.shape[0], Y.shape[1]))
Yreal[0::2] = np.real(Y)
Yreal[1::2] = np.imag(Y)
return Yreal
def cost(self, controls, densities, system_eval_step):
"""
Compute the penalty.
Arguments:
controls
densities
system_eval_step
Returns:
cost
"""
# The cost is the overlap (fidelity) of the evolved density and each
# forbidden density.
cost = 0
for i, forbidden_densities_dagger_ in enumerate(
self.forbidden_densities_dagger):
density = densities[i]
density_cost = 0
for forbidden_density_dagger in forbidden_densities_dagger_:
inner_product = (
anp.trace(anp.matmul(forbidden_density_dagger, density)) /
self.hilbert_size)
fidelity = anp.real(inner_product *
anp.conjugate(inner_product))
density_cost = density_cost + fidelity
#ENDFOR
density_cost_normalized = density_cost / self.forbidden_densities_count[
i]
cost = cost + density_cost_normalized
#ENDFOR
# Normalize the cost for the number of evolving densities
# and the number of times the cost is computed.
cost_normalized = cost / self.cost_normalization_constant
return cost_normalized * self.cost_multiplier
def test_real_type():
fun = lambda x: np.sum(np.real(x))
df = grad(fun)
assert type(df(1.0)) == float
assert type(df(1.0j)) == complex
def u2r(*tlist1):
return np.real(u2_alt(*tlist1))
Returns
-------
np.ndarray
shape=(2,2)
"""
# print('mmbbvv, pu2', pu2r(*tlist) +1j* pu2r(*tlist))
return pu2r(*tlist) + 1j * pu2i(*tlist)
defvjp(
pu2r,
# defines vector-jacobian-product of pu2r
# g.shape == pu2r.shape
lambda ans, *tlist: lambda g: np.sum(g * np.real(d_u2(0, *tlist))),
lambda ans, *tlist: lambda g: np.sum(g * np.real(d_u2(1, *tlist))),
lambda ans, *tlist: lambda g: np.sum(g * np.real(d_u2(2, *tlist))),
lambda ans, *tlist: lambda g: np.sum(g * np.real(d_u2(3, *tlist))),
argnums=range(4))
defvjp(
pu2i,
# defines vector-jacobian-product of pu2i
# g.shape == pu2i.shape
lambda ans, *tlist: lambda g: np.sum(g * np.imag(d_u2(0, *tlist))),
lambda ans, *tlist: lambda g: np.sum(g * np.imag(d_u2(1, *tlist))),
lambda ans, *tlist: lambda g: np.sum(g * np.imag(d_u2(2, *tlist))),
lambda ans, *tlist: lambda g: np.sum(g * np.imag(d_u2(3, *tlist))),
argnums=range(4))
def to_scalar(x):
if isinstance(getval(x), list) or isinstance(getval(x), tuple):
return sum([to_scalar(item) for item in x])
return np.sum(np.real(np.sin(x)))
def test_real_type():
fun = lambda x: np.sum(np.real(x))
df = grad(fun)
assert np.isrealobj(df(2.0))
assert np.iscomplexobj(df(1.0j))
def norm(x):
return np.mean(np.square(np.real(x)) + np.square(np.imag(x)))
def to_scalar(x):
if isinstance(getval(x), list) or isinstance(getval(x), tuple):
return sum([to_scalar(item) for item in x])
return np.sum(np.real(np.sin(x)))
#### find R_max via bisection method #### # min value T T1=1e-8 shannon_capacity_fact_fixed_time = make_shannon_capacity_fact_fixed_time(T1) problem = Problem(manifold=manifold_fact, cost=shannon_capacity_fact_fixed_time, verbosity=0) L_opt1 = solver.solve(problem) p_1 = -shannon_capacity_fact_fixed_time(L_opt1) #print p_1 Sigma_opt1 = np.dot(L_opt1,L_opt1.transpose()) Sigma_opt_norm1 = Sigma_opt1/(np.trace(Sigma_opt1)) eigs, eigvs = np.linalg.eig(Sigma_opt_norm1) eff_rank1 = np.square(np.sum(np.real(eigs)))/np.sum(np.square(np.real(eigs))) # max value T T2=10. shannon_capacity_fact_fixed_time = make_shannon_capacity_fact_fixed_time(T2) problem = Problem(manifold=manifold_fact, cost=shannon_capacity_fact_fixed_time, verbosity=0) L_opt2 = solver.solve(problem) p_2 = -shannon_capacity_fact_fixed_time(L_opt2) #print p_2 # mean value T Tm = (T1+T2)/2 shannon_capacity_fact_fixed_time = make_shannon_capacity_fact_fixed_time(Tm) problem = Problem(manifold=manifold_fact, cost=shannon_capacity_fact_fixed_time, verbosity=0) L_optm = solver.solve(problem) p_m = -shannon_capacity_fact_fixed_time(L_optm)
def shift_spectrum(self, spec, shift_in_pxl):
kernel = np.exp(-1j*2*np.pi*self.freq_grid*shift_in_pxl)
return np.real(np.fft.ifft( np.fft.fft(spec) * kernel ))
def vjp(g):
return anp.real(solve_sylvester(ans_transp, ans_transp, g))
def fun(a):
r, i = np.real(a), np.imag(a)
a = np.abs(r)**1.4 + np.abs(i)**1.3
return np.sum(np.sin(a))
def holomorphic_grad(fun, x):
if not vspace(x).iscomplex:
warnings.warn("Input to holomorphic_grad is not complex")
return grad(lambda x: np.real(fun(x)))(x)
for pp in range(M):
# upload C. elegans connectivity matrix (insert full file path)
A = np.loadtxt("insert_full_path/A_C_elegans.txt", usecols=range(n))
for tt in range(279):
for qq in range(279):
if tt > qq:
if random() < 0.5:
A_tmp = A[tt, qq]
A[tt, qq] = A[qq, tt]
A[qq, tt] = A_tmp
w, v = np.linalg.eig(A)
A = A - (max(np.real(w)) + 0.1) * I
# vector of transmission rates
rate_vec = np.zeros(len(T_vec))
k = 0
# plot Shannon transmission rate R_T over time
for T in T_vec:
### compute Shannon transmission rate ###
def dlyap_iterative(a, q, eps=LYAPUNOV_EPSILON, iter_limit=ITER_LIMIT):
error = 1E+6
x = q
Nr = TrackNormal(rx)
# psie is sort of backwards: higher angles go to the left
return np.angle(Nx) - np.angle(Nr)
if __name__ == '__main__':
TRACK_SPACING = 19.8 # cm
x = SVGPathToTrackPoints("oakwarehouse.path", TRACK_SPACING)[:-1]
xm = np.array(x)[:, 0] / 50 # 50 pixels / meter
track_k = TrackCurvature(xm)
Nx = TrackNormal(xm)
u = 1j * Nx
np.savetxt("track_x.txt",
np.vstack([np.real(xm), np.imag(xm)]).T.reshape(-1),
newline=",\n")
np.savetxt("track_u.txt",
np.vstack([np.real(u), np.imag(u)]).T.reshape(-1),
newline=",\n")
np.savetxt("track_k.txt", track_k, newline=",\n")
ye, val, stuff = OptimizeTrack(xm, 1.4, 0.1)
psie = RelativePsie(ye, xm)
rx = u*ye + xm
raceline_k = TrackCurvature(rx)
np.savetxt("raceline_k.txt", raceline_k, newline=",\n")
np.savetxt("raceline_ye.txt", ye, newline=",\n")
np.savetxt("raceline_psie.txt", psie, newline=",\n")
Nr = TrackNormal(rx)
# psie is sort of backwards: higher angles go to the left
return np.angle(Nx) - np.angle(Nr)
if __name__ == '__main__':
TRACK_SPACING = 19.8 # cm
x = SVGPathToTrackPoints("oakwarehouse.path", TRACK_SPACING)[:-1]
xm = np.array(x)[:, 0] / 50 # 50 pixels / meter
track_k = TrackCurvature(xm)
Nx = TrackNormal(xm)
u = 1j * Nx
np.savetxt("track_x.txt",
np.vstack([np.real(xm), np.imag(xm)]).T.reshape(-1),
newline=",\n")
np.savetxt("track_u.txt",
np.vstack([np.real(u), np.imag(u)]).T.reshape(-1),
newline=",\n")
np.savetxt("track_k.txt", track_k, newline=",\n")
ye, val, stuff = OptimizeTrack(xm, 1.4, 0.1)
psie = RelativePsie(ye, xm)
rx = u * ye + xm
raceline_k = TrackCurvature(rx)
np.savetxt("raceline_k.txt", raceline_k, newline=",\n")
np.savetxt("raceline_ye.txt", ye, newline=",\n")
np.savetxt("raceline_psie.txt", psie, newline=",\n")
def polyinterp(points, doPlot=None, xminBound=None, xmaxBound=None):
""" polynomial interpolation
Parameters
----------
points: shape(pointNum, 3), three columns represents x, f, g
doPolot: set to 1 to plot, default 0
xmin: min value that brackets minimum (default: min of points)
xmax: max value that brackets maximum (default: max of points)
set f or g to sqrt(-1)=1j if they are not known
the order of the polynomial is the number of known f and g values minus 1
Returns
-------
minPos:
fmin:
"""
if doPlot == None:
doPlot = 0
nPoints = points.shape[0]
order = np.sum(np.imag(points[:, 1:3]) == 0) -1
# code for most common case: cubic interpolation of 2 points
if nPoints == 2 and order == 3 and doPlot == 0:
[minVal, minPos] = [np.min(points[:,0]), np.argmin(points[:,0])]
notMinPos = 1 - minPos
d1 = points[minPos,2] + points[notMinPos,2] - 3*(points[minPos,1]-\
points[notMinPos,1])/(points[minPos,0]-points[notMinPos,0])
t_d2 = d1**2 - points[minPos,2]*points[notMinPos,2]
if t_d2 > 0:
d2 = np.sqrt(t_d2)
else:
d2 = np.sqrt(-t_d2) * np.complex(0,1)
if np.isreal(d2):
t = points[notMinPos,0] - (points[notMinPos,0]-points[minPos,0])*\
((points[notMinPos,2]+d2-d1)/(points[notMinPos,2]-\
points[minPos,2]+2*d2))
minPos = np.min([np.max([t,points[minPos,0]]), points[notMinPos,0]])
else:
minPos = np.mean(points[:,0])
fmin = minVal
return (minPos, fmin)
xmin = np.min(points[:,0])
xmax = np.max(points[:,0])
# compute bounds of interpolation area
if xminBound == None:
xminBound = xmin
if xmaxBound == None:
xmaxBound = xmax
# constraints based on available function values
A = np.zeros((0, order+1))
b = np.zeros((0, 1))
for i in range(nPoints):
if np.imag(points[i,1]) == 0:
constraint = np.zeros(order+1)
for j in np.arange(order,-1,-1):
constraint[order-j] = points[i,0]**j
A = np.vstack((A, constraint))
b = np.append(b, points[i,1])
# constraints based on availabe derivatives
for i in range(nPoints):
if np.isreal(points[i,2]):
constraint = np.zeros(order+1)
for j in range(1,order+1):
constraint[j-1] = (order-j+1)* points[i,0]**(order-j)
A = np.vstack((A, constraint))
b = np.append(b,points[i,2])
# find interpolating polynomial
params = np.linalg.solve(A, b)
# compute critical points
dParams = np.zeros(order)
for i in range(params.size-1):
dParams[i] = params[i] * (order-i)
if np.any(np.isinf(dParams)):
cp = np.concatenate((np.array([xminBound, xmaxBound]), points[:,0]))
else:
cp = np.concatenate((np.array([xminBound, xmaxBound]), points[:,0], \
np.roots(dParams)))
# test critical points
fmin = np.infty;
minPos = (xminBound + xmaxBound)/2.
for xCP in cp:
if np.imag(xCP) == 0 and xCP >= xminBound and xCP <= xmaxBound:
fCP = np.polyval(params, xCP)
if np.imag(fCP) == 0 and fCP < fmin:
minPos = np.double(np.real(xCP))
fmin = np.double(np.real(fCP))
# plot situation (omit this part for now since we are not going to use it
# anyway)
return (minPos, fmin)
def fun(a):
b = a + 1.0j
c = b[0] + 1.5
d = a + b
e = d + c
return np.sum(np.sin(np.real(e)))
def fun(a):
r, i = np.real(a), np.imag(a)
a = np.abs(r)**1.4 + np.abs(i)**1.3
return to_scalar(a)
def to_scalar(x):
return np.sum(np.real(np.sin(x)))
def to_scalar(x):
if isinstance(x, list) or isinstance(x, ListNode) or \
isinstance(x, tuple) or isinstance(x, TupleNode):
return sum([to_scalar(item) for item in x])
return np.sum(np.real(np.sin(x)))
def test_real_type():
fun = lambda x: np.sum(np.real(x))
df = grad(fun)
assert type(df(1.0)) == float
assert type(df(1.0j)) == complex
def test_falseyness():
fun = lambda x: np.real(x**2 if np.iscomplex(x) else np.sum(x))
check_grads(fun)(2.)
check_grads(fun)(2. + 1j)
