def __init__(self,
phase_obj_3d,
wavelength,
slice_binning_factor=1,
**kwargs):
"""
Initialization of the class
phase_obj_3d: object of the class PhaseObject3D
wavelength: wavelength of the light
slice_binning_factor: the object is compress in z-direction by this factor
"""
self.slice_binning_factor = slice_binning_factor
self._shape_full = phase_obj_3d.shape
self.shape = phase_obj_3d.shape[0:2] + (int(
np.ceil(phase_obj_3d.shape[2] / self.slice_binning_factor)), )
self.RI = phase_obj_3d.RI
self.wavelength = wavelength
self.pixel_size = phase_obj_3d.pixel_size
self.pixel_size_z = phase_obj_3d.pixel_size_z * self.slice_binning_factor
self.back_scatter = False
#Broadcasts b to a, size(a) > size(b)
self.assign_broadcast = lambda a, b: a - a + b
fxlin, fylin = self._genFrequencyGrid()
self.fzlin = ((self.RI / self.wavelength)**2 -
fxlin * af.conjg(fxlin) - fylin * af.conjg(fylin))**0.5
self.prop_kernel_phase = 1.0j * 2.0 * np.pi * self.fzlin
def adjoint(self, residual, cache):
V_obj_af, field_layer_in_or_grad,\
flag_gpu_inout = cache
#back-propagte to volume center
field_bp = residual
if self.focus_at_center:
#propagate to the last layer
field_bp = self._propagationInplace(field_bp,
self.distance_end_to_center)
for layer in range(self.shape[2] - 1, -1, -1):
field_bp_1 = af.ifft2(
af.fft2(field_bp) *
af.conjg(self.green_kernel_2d)) * self.pixel_size_z
if layer > 0:
field_bp = self._propagationInplace(
field_bp, self.slice_separation[layer], adjoint=True)
field_bp[:, :] += field_bp_1 * af.conjg(V_obj_af[:, :, layer])
field_layer_in_or_grad[:, :, layer] = field_bp_1 * af.conjg(
field_layer_in_or_grad[:, :, layer])
#Unbinning
grad = self._meanObject(field_layer_in_or_grad, adjoint=True)
if flag_gpu_inout:
return {'gradient': grad}
else:
return {'gradient': np.array(grad)}
def __init__(self,
shape,
pixel_size,
wavelength,
na,
RI_measure=1.0,
**kwargs):
"""
Initialization of the class
RI_measure: refractive index on the detection side (example: oil immersion objectives)
"""
super().__init__(shape, pixel_size, na, **kwargs)
fxlin = genGrid(self.shape[1],
1.0 / self.pixel_size / self.shape[1],
flag_shift=True)
fylin = genGrid(self.shape[0],
1.0 / self.pixel_size / self.shape[0],
flag_shift=True)
fxlin = af.tile(fxlin.T, self.shape[0], 1)
fylin = af.tile(fylin, 1, self.shape[1])
self.pupil_support = genPupil(self.shape, self.pixel_size, self.na,
wavelength)
self.prop_kernel_phase = 1.0j * 2.0 * np.pi * self.pupil_support * (
(RI_measure / wavelength)**2 - fxlin * af.conjg(fxlin) -
fylin * af.conjg(fylin))**0.5
def _split_fourier_cuda(self, signal: Signal, step):
'''
This function is called by split_fourier,and should not be used outside
:param signal: signal to traverse the span
:param step: the step of split fourier
:return: None
'''
af.set_backend('cuda')
freq = fftfreq(len(signal.data_sample[0, :]),
(signal.sps * signal.symbol_rate_in_hz)**(-1))
freq = af.Array(freq.ctypes.data, freq.shape, freq.dtype.char)
signal_x = np.asarray(signal.data_sample[0, :])
signal_y = np.asarray(signal.data_sample[1, :])
signal_x = af.Array(signal_x.ctypes.data,
signal_x.shape,
dtype=signal_x.dtype.char)
signal_y = af.Array(signal_y.ctypes.data,
signal_x.shape,
dtype=signal_y.dtype.char)
Disper = (1j / 2) * self.beta2 * (2 * np.pi * freq)**2 * step + (
1j / 6) * self.beta3 * (
(2 * np.pi * freq)**3 * step) - self.alphalin / 2 * step
dz_Eff = (1 - np.exp(-self.alphalin * step)) / self.alphalin
step_number = np.ceil(self.length / step)
for number in range(int(step_number)):
print(number)
if number == step_number - 1:
# dz = step
dz = self.length - (step_number - 1) * step
dz_Eff = (1 - np.exp(-self.alphalin * dz)) / self.alphalin
Disper = (1j / 2) * self.beta2 * (2 * np.pi * freq)**2 * dz + (
1j / 6) * self.beta3 * (
(2 * np.pi * freq)**3 * dz) - self.alphalin / 2 * step
signal_x, signal_y = self.linear(signal_x, signal_y, Disper)
energy = signal_x * af.conjg(signal_x) + signal_y * af.conjg(
signal_y)
signal_x, signal_y = self.nonlinear(energy, signal_x, signal_y,
dz_Eff)
signal_x, signal_y = self.linear(signal_x, signal_y, Disper)
signal_x_array = np.array(signal_x.to_list())
signal_y_array = np.array(signal_y.to_list())
signal_x_array = signal_x_array[:, 0] + 1j * signal_x_array[:, 1]
signal_y_array = signal_y_array[:, 0] + 1j * signal_y_array[:, 1]
signal.data_sample[0, :] = signal_x_array
signal.data_sample[1, :] = signal_y_array
def func(x0):
fields = self._scattering_obj.forward(x0, fx_illu, fy_illu)
field_scattered = self._defocus_obj.forward(field_scattered, self.prop_distances)
field_measure = self._crop_obj.forward(field_scattered)
residual = af.abs(field_measure) - amplitude
function_value = af.sum(residual*af.conjg(residual)).real
return function_value
def adjoint(self, residual, cache):
phasecontrast_obj_af, field_layer_conj_or_grad, flag_gpu_inout = cache
trans_obj_af_conj = af.conjg(phasecontrast_obj_af)
#back-propagte to volume center
field_bp = residual
#propagate to the last layer
if self.focus_at_center:
field_bp = self._propagationInplace(field_bp,
self.distance_end_to_center)
#multi-slice transmittance backward
for layer in range(self.shape[2] - 1, -1, -1):
field_layer_conj_or_grad[:, :,
layer] = field_bp * field_layer_conj_or_grad[:, :, layer] * (
-1.0j
) * self.sigma * trans_obj_af_conj[:, :,
layer]
if layer > 0:
field_bp[:, :] *= trans_obj_af_conj[:, :, layer]
field_bp = self._propagationInplace(
field_bp, self.slice_separation[layer - 1], adjoint=True)
#Unbinning
grad = self._binObject(field_layer_conj_or_grad, adjoint=True)
phasecontrast_obj_af = None
if flag_gpu_inout:
return {'gradient': grad}
else:
return {'gradient': np.array(grad)}
def conj(self):
if not numpy.issubdtype(self.dtype, numpy.complex):
return afnumpy.copy(self)
if(self.d_array is not None):
s = arrayfire.conjg(self.d_array)
return ndarray(self.shape, dtype=pu.typemap(s.dtype()), af_array=s)
else:
return self.h_array.conj()
def adjoint(self, field):
"""Adjoint operator for pupil (and estimate pupil if selected)"""
field_adj_f = af.fft2(field)
field_pupil_adj = af.ifft2(af.conjg(self.getPupil()) * field_adj_f)
#Pupil recovery
if self.flag_update:
self._update(field_adj_f)
return field_pupil_adj
def forward(self, field, **kwargs):
"""Apply pupil"""
self.field_f = af.fft2(field)
if self.flag_update:
if self.update_method == "GaussNewton":
self.approx_hessian[:, :] += self.field_f * af.conjg(
self.field_f)
field = af.ifft2(self.getPupil() * self.field_f)
return field
def __init__(self, phase_obj_3d, wavelength, **kwargs):
super().__init__(phase_obj_3d, wavelength, **kwargs)
self.distance_end_to_center = np.sum(
self.slice_separation) / 2. + phase_obj_3d.slice_separation[-1]
self.distance_beginning_to_center = np.sum(self.slice_separation) / 2.
self.slice_separation = np.append(phase_obj_3d.slice_separation,
[phase_obj_3d.slice_separation[-1]])
self.slice_separation = np.asarray([sum(self.slice_separation[x:x+self.slice_binning_factor]) \
for x in range(0,len(self.slice_separation),self.slice_binning_factor)])
if self.slice_separation.shape[0] != self.shape[2]:
print(
"Number of slices does not match with number of separations!")
# generate green's function convolution kernel
fxlin, fylin = self._genFrequencyGrid()
kernel_mask = (self.RI / self.wavelength)**2 > 1.01 * (
fxlin * af.conjg(fxlin) + fylin * af.conjg(fylin))
self.green_kernel_2d = -0.25j * af.exp(
2.0j * np.pi * self.fzlin * self.pixel_size_z) / np.pi / self.fzlin
self.green_kernel_2d *= kernel_mask
self.green_kernel_2d[af.isnan(self.green_kernel_2d) == 1] = 0.0
def forward(self, contrast_obj, fx_illu, fy_illu):
#compute illumination
with contexttimer.Timer() as timer:
field, fx_illu, fy_illu, fz_illu = self._genIllumination(
fx_illu, fy_illu)
field[:, :] *= np.exp(1.0j * 2.0 * np.pi * fz_illu *
self.initial_z_position)
field_layer_conj = af.constant(0.0,
self.shape[0],
self.shape[1],
self.shape[2],
dtype=af_complex_datatype)
if (type(contrast_obj).__module__ == np.__name__):
phasecontrast_obj_af = af.to_array(contrast_obj)
flag_gpu_inout = False
else:
phasecontrast_obj_af = contrast_obj
flag_gpu_inout = True
#Binning
obj_af = self._binObject(phasecontrast_obj_af)
#Potentials to Transmittance
obj_af = af.exp(1.0j * self.sigma * obj_af)
for layer in range(self.shape[2]):
field_layer_conj[:, :, layer] = af.conjg(field)
field[:, :] *= obj_af[:, :, layer]
if layer < self.shape[2] - 1:
field = self._propagationInplace(
field, self.slice_separation[layer])
#propagate to volume center
cache = (obj_af, field_layer_conj, flag_gpu_inout)
if self.focus_at_center:
#propagate to volume center
field = self._propagationInplace(field,
self.distance_end_to_center,
adjoint=True)
return {'forward_scattered_field': field, 'cache': cache}
def _update(self, field_adj_f):
"""function to recover pupil"""
self.pupil_gradient_phase[:, :] += af.conjg(self.field_f) * field_adj_f
self.measure_count += 1
if self.measure_count == self.measurement_num:
if self.update_method == "gradient":
self.pupil[:, :] -= self.pupil_step_size * self.pupil_gradient_phase * self.pupil_support
elif self.update_method == "GaussNewton":
self.pupil[:, :] -= self.pupil_step_size * af.abs(
self.field_f
) * self.pupil_gradient_phase * self.pupil_support / (
(self.approx_hessian + 1e-3) *
af.max(af.abs(self.field_f[:])))
self.approx_hessian[:, :] = 0.0
else:
print("there is no update_method \"%s\"!" %
(self.update_method))
raise
self.pupil_gradient_phase[:, :] = 0.0
self.measure_count = 0
def forward(self, trans_obj, fx_illu, fy_illu):
if self.slice_binning_factor > 1:
print(
"Slicing is not implemented for MultiTransmittance algorithm!")
raise
#compute illumination
field, fx_illu, fy_illu, fz_illu = self._genIllumination(
fx_illu, fy_illu)
field[:, :] *= np.exp(1.0j * 2.0 * np.pi * fz_illu *
self.initial_z_position)
#multi-slice transmittance forward propagation
field_layer_conj = af.constant(0.0,
trans_obj.shape[0],
trans_obj.shape[1],
trans_obj.shape[2],
dtype=af_complex_datatype)
if (type(trans_obj).__module__ == np.__name__):
trans_obj_af = af.to_array(trans_obj)
flag_gpu_inout = False
else:
trans_obj_af = trans_obj
flag_gpu_inout = True
for layer in range(self.shape[2]):
field_layer_conj[:, :, layer] = af.conjg(field)
field[:, :] *= trans_obj_af[:, :, layer]
if layer < self.shape[2] - 1:
field = self._propagationInplace(field,
self.slice_separation[layer])
#store intermediate variables for adjoint operation
cache = (trans_obj_af, field_layer_conj, flag_gpu_inout)
if self.focus_at_center:
#propagate to volume center
field = self._propagationInplace(field,
self.distance_end_to_center,
adjoint=True)
return {'forward_scattered_field': field, 'cache': cache}
def propKernel(shape,
pixel_size,
wavelength,
prop_distance,
NA=None,
RI=1.0,
band_limited=True):
assert len(shape) == 2, "pupil should be two dimensional!"
fxlin = genGrid(shape[1], 1 / pixel_size / shape[1], flag_shift=True)
fylin = genGrid(shape[0], 1 / pixel_size / shape[0], flag_shift=True)
fxlin = af.tile(fxlin.T, shape[0], 1)
fylin = af.tile(fylin, 1, shape[1])
if band_limited:
assert NA is not None, "need to provide numerical aperture of the system!"
Pcrop = genPupil(shape, pixel_size, NA, wavelength)
else:
Pcrop = 1.0
prop_kernel = Pcrop * af.exp(1.0j * 2.0 * np.pi * abs(prop_distance) * Pcrop *\
((RI/wavelength)**2 - fxlin**2 - fylin**2)**0.5)
prop_kernel = af.conjg(prop_kernel) if prop_distance < 0 else prop_kernel
return prop_kernel
def adjoint(self, residual, cache):
trans_obj_af, field_layer_conj_or_grad, flag_gpu_inout = cache
#back-propagte to volume center
field_bp = residual
if self.focus_at_center:
#propagate to the last layer
field_bp = self._propagationInplace(field_bp,
self.distance_end_to_center)
#multi-slice transmittance backward
for layer in range(self.shape[2] - 1, -1, -1):
field_layer_conj_or_grad[:, :,
layer] = field_bp * field_layer_conj_or_grad[:, :,
layer]
if layer > 0:
field_bp[:, :] *= af.conjg(trans_obj_af[:, :, layer])
field_bp = self._propagationInplace(
field_bp, self.slice_separation[layer - 1], adjoint=True)
if flag_gpu_inout:
return {'gradient': field_layer_conj_or_grad}
else:
return {'gradient': np.array(field_layer_conj_or_grad)}
def _propagationInplace(self,
field,
propagation_distance,
adjoint=False,
in_real=True):
"""
propagation operator that uses angular spectrum to propagate the wave
field: input field
propagation_distance: distance to propagate the wave
adjoint: boolean variable to perform adjoint operation (i.e. opposite direction)
"""
if in_real:
af.fft2_inplace(field)
if adjoint:
field[:, :] *= af.conjg(
af.exp(self.prop_kernel_phase * propagation_distance))
else:
field[:, :] *= af.exp(self.prop_kernel_phase *
propagation_distance)
if in_real:
af.ifft2_inplace(field)
return field
def simple_arith(verbose=False):
display_func = _util.display_func(verbose)
print_func = _util.print_func(verbose)
a = af.randu(3, 3)
b = af.constant(4, 3, 3)
display_func(a)
display_func(b)
c = a + b
d = a
d += b
display_func(c)
display_func(d)
display_func(a + 2)
display_func(3 + a)
c = a - b
d = a
d -= b
display_func(c)
display_func(d)
display_func(a - 2)
display_func(3 - a)
c = a * b
d = a
d *= b
display_func(c * 2)
display_func(3 * d)
display_func(a * 2)
display_func(3 * a)
c = a / b
d = a
d /= b
display_func(c / 2.0)
display_func(3.0 / d)
display_func(a / 2)
display_func(3 / a)
c = a % b
d = a
d %= b
display_func(c % 2.0)
display_func(3.0 % d)
display_func(a % 2)
display_func(3 % a)
c = a**b
d = a
d **= b
display_func(c**2.0)
display_func(3.0**d)
display_func(a**2)
display_func(3**a)
display_func(a < b)
display_func(a < 0.5)
display_func(0.5 < a)
display_func(a <= b)
display_func(a <= 0.5)
display_func(0.5 <= a)
display_func(a > b)
display_func(a > 0.5)
display_func(0.5 > a)
display_func(a >= b)
display_func(a >= 0.5)
display_func(0.5 >= a)
display_func(a != b)
display_func(a != 0.5)
display_func(0.5 != a)
display_func(a == b)
display_func(a == 0.5)
display_func(0.5 == a)
a = af.randu(3, 3, dtype=af.Dtype.u32)
b = af.constant(4, 3, 3, dtype=af.Dtype.u32)
display_func(a & b)
display_func(a & 2)
c = a
c &= 2
display_func(c)
display_func(a | b)
display_func(a | 2)
c = a
c |= 2
display_func(c)
display_func(a >> b)
display_func(a >> 2)
c = a
c >>= 2
display_func(c)
display_func(a << b)
display_func(a << 2)
c = a
c <<= 2
display_func(c)
display_func(-a)
display_func(+a)
display_func(~a)
display_func(a)
display_func(af.cast(a, af.Dtype.c32))
display_func(af.maxof(a, b))
display_func(af.minof(a, b))
display_func(af.rem(a, b))
a = af.randu(3, 3) - 0.5
b = af.randu(3, 3) - 0.5
display_func(af.abs(a))
display_func(af.arg(a))
display_func(af.sign(a))
display_func(af.round(a))
display_func(af.trunc(a))
display_func(af.floor(a))
display_func(af.ceil(a))
display_func(af.hypot(a, b))
display_func(af.sin(a))
display_func(af.cos(a))
display_func(af.tan(a))
display_func(af.asin(a))
display_func(af.acos(a))
display_func(af.atan(a))
display_func(af.atan2(a, b))
c = af.cplx(a)
d = af.cplx(a, b)
display_func(c)
display_func(d)
display_func(af.real(d))
display_func(af.imag(d))
display_func(af.conjg(d))
display_func(af.sinh(a))
display_func(af.cosh(a))
display_func(af.tanh(a))
display_func(af.asinh(a))
display_func(af.acosh(a))
display_func(af.atanh(a))
a = af.abs(a)
b = af.abs(b)
display_func(af.root(a, b))
display_func(af.pow(a, b))
display_func(af.pow2(a))
display_func(af.sigmoid(a))
display_func(af.exp(a))
display_func(af.expm1(a))
display_func(af.erf(a))
display_func(af.erfc(a))
display_func(af.log(a))
display_func(af.log1p(a))
display_func(af.log10(a))
display_func(af.log2(a))
display_func(af.sqrt(a))
display_func(af.cbrt(a))
a = af.round(5 * af.randu(3, 3) - 1)
b = af.round(5 * af.randu(3, 3) - 1)
display_func(af.factorial(a))
display_func(af.tgamma(a))
display_func(af.lgamma(a))
display_func(af.iszero(a))
display_func(af.isinf(a / b))
display_func(af.isnan(a / a))
a = af.randu(5, 1)
b = af.randu(1, 5)
c = af.broadcast(lambda x, y: x + y, a, b)
display_func(a)
display_func(b)
display_func(c)
@af.broadcast
def test_add(aa, bb):
return aa + bb
display_func(test_add(a, b))
def conj(self):
if not numpy.issubdtype(self.dtype, numpy.complex):
return afnumpy.copy(self)
s = arrayfire.conjg(self.d_array)
return ndarray(self.shape, dtype=pu.typemap(s.dtype()), af_array=s)
def cong(arr):
return af.conjg(arr)
af.display(af.hypot(a, b)) af.display(af.sin(a)) af.display(af.cos(a)) af.display(af.tan(a)) af.display(af.asin(a)) af.display(af.acos(a)) af.display(af.atan(a)) af.display(af.atan2(a, b)) c = af.cplx(a) d = af.cplx(a, b) af.display(c) af.display(d) af.display(af.real(d)) af.display(af.imag(d)) af.display(af.conjg(d)) af.display(af.sinh(a)) af.display(af.cosh(a)) af.display(af.tanh(a)) af.display(af.asinh(a)) af.display(af.acosh(a)) af.display(af.atanh(a)) a = af.abs(a) b = af.abs(b) af.display(af.root(a, b)) af.display(af.pow(a, b)) af.display(af.pow2(a)) af.display(af.exp(a))
def cart2Pol(x, y):
rho = (x * af.conjg(x) + y * af.conjg(y))**0.5
theta = af.atan2(af.real(y), af.real(x)).as_type(af_complex_datatype)
return rho, theta
def vdot(a, b):
s = arrayfire.dot(arrayfire.conjg(a.flat.d_array), b.flat.d_array)
return afnumpy.ndarray((), dtype=a.dtype, af_array=s)[()]
def simple_arith(verbose = False):
display_func = _util.display_func(verbose)
print_func = _util.print_func(verbose)
a = af.randu(3,3,dtype=af.Dtype.u32)
b = af.constant(4, 3, 3, dtype=af.Dtype.u32)
display_func(a)
display_func(b)
c = a + b
d = a
d += b
display_func(c)
display_func(d)
display_func(a + 2)
display_func(3 + a)
c = a - b
d = a
d -= b
display_func(c)
display_func(d)
display_func(a - 2)
display_func(3 - a)
c = a * b
d = a
d *= b
display_func(c * 2)
display_func(3 * d)
display_func(a * 2)
display_func(3 * a)
c = a / b
d = a
d /= b
display_func(c / 2.0)
display_func(3.0 / d)
display_func(a / 2)
display_func(3 / a)
c = a % b
d = a
d %= b
display_func(c % 2.0)
display_func(3.0 % d)
display_func(a % 2)
display_func(3 % a)
c = a ** b
d = a
d **= b
display_func(c ** 2.0)
display_func(3.0 ** d)
display_func(a ** 2)
display_func(3 ** a)
display_func(a < b)
display_func(a < 0.5)
display_func(0.5 < a)
display_func(a <= b)
display_func(a <= 0.5)
display_func(0.5 <= a)
display_func(a > b)
display_func(a > 0.5)
display_func(0.5 > a)
display_func(a >= b)
display_func(a >= 0.5)
display_func(0.5 >= a)
display_func(a != b)
display_func(a != 0.5)
display_func(0.5 != a)
display_func(a == b)
display_func(a == 0.5)
display_func(0.5 == a)
display_func(a & b)
display_func(a & 2)
c = a
c &= 2
display_func(c)
display_func(a | b)
display_func(a | 2)
c = a
c |= 2
display_func(c)
display_func(a >> b)
display_func(a >> 2)
c = a
c >>= 2
display_func(c)
display_func(a << b)
display_func(a << 2)
c = a
c <<= 2
display_func(c)
display_func(-a)
display_func(+a)
display_func(~a)
display_func(a)
display_func(af.cast(a, af.Dtype.c32))
display_func(af.maxof(a,b))
display_func(af.minof(a,b))
display_func(af.rem(a,b))
a = af.randu(3,3) - 0.5
b = af.randu(3,3) - 0.5
display_func(af.abs(a))
display_func(af.arg(a))
display_func(af.sign(a))
display_func(af.round(a))
display_func(af.trunc(a))
display_func(af.floor(a))
display_func(af.ceil(a))
display_func(af.hypot(a, b))
display_func(af.sin(a))
display_func(af.cos(a))
display_func(af.tan(a))
display_func(af.asin(a))
display_func(af.acos(a))
display_func(af.atan(a))
display_func(af.atan2(a, b))
c = af.cplx(a)
d = af.cplx(a,b)
display_func(c)
display_func(d)
display_func(af.real(d))
display_func(af.imag(d))
display_func(af.conjg(d))
display_func(af.sinh(a))
display_func(af.cosh(a))
display_func(af.tanh(a))
display_func(af.asinh(a))
display_func(af.acosh(a))
display_func(af.atanh(a))
a = af.abs(a)
b = af.abs(b)
display_func(af.root(a, b))
display_func(af.pow(a, b))
display_func(af.pow2(a))
display_func(af.exp(a))
display_func(af.expm1(a))
display_func(af.erf(a))
display_func(af.erfc(a))
display_func(af.log(a))
display_func(af.log1p(a))
display_func(af.log10(a))
display_func(af.log2(a))
display_func(af.sqrt(a))
display_func(af.cbrt(a))
a = af.round(5 * af.randu(3,3) - 1)
b = af.round(5 * af.randu(3,3) - 1)
display_func(af.factorial(a))
display_func(af.tgamma(a))
display_func(af.lgamma(a))
display_func(af.iszero(a))
display_func(af.isinf(a/b))
display_func(af.isnan(a/a))
a = af.randu(5, 1)
b = af.randu(1, 5)
c = af.broadcast(lambda x,y: x+y, a, b)
display_func(a)
display_func(b)
display_func(c)
@af.broadcast
def test_add(aa, bb):
return aa + bb
display_func(test_add(a, b))
def _solveFirstOrderGradient(self, measurements, verbose, callback=None):
"""
MAIN part of the solver, runs the FISTA algorithm
configs: configs object from class AlgorithmConfigs
measurements: all measurements
self.configs.recon_from_field == True: field
self.configs.recon_from_field == False: amplitude measurement
verbose: boolean variable to print verbosely
"""
flag_FISTA = False
if self.configs.method == "FISTA":
flag_FISTA = True
# update multiple angles at a time
batch_update = False
if self.configs.fista_global_update or self.configs.batch_size != 1:
gradient_batch = af.constant(0.0, self.phase_obj_3d.shape[0],\
self.phase_obj_3d.shape[1],\
self.phase_obj_3d.shape[2], dtype = af_complex_datatype)
batch_update = True
if self.configs.fista_global_update:
self.configs.batch_size = 0
#TODO: what if num_batch is not an integer
if self.configs.batch_size == 0:
num_batch = 1
else:
num_batch = self.number_illum // self.configs.batch_size
stepsize = self.configs.stepsize
max_iter = self.configs.max_iter
reg_term = self.configs.reg_term
self.configs.error = []
obj_gpu = af.constant(0.0, self.phase_obj_3d.shape[0],\
self.phase_obj_3d.shape[1],\
self.phase_obj_3d.shape[2], dtype = af_complex_datatype)
#Initialization for FISTA update
if flag_FISTA:
restart = self.configs.restart
y_k = self._x.copy()
t_k = 1.0
#Set Callback flag
if callback is None:
run_callback = False
else:
run_callback = True
#Start of iterative algorithm
with contexttimer.Timer() as timer:
if verbose:
print("---- Start of the %5s algorithm ----" %(self.scat_model))
for iteration in range(max_iter):
illu_counter = 0
cost = 0.0
obj_gpu[:] = af.to_array(self._x)
if self.configs.random_order:
illu_order = np.random.permutation(range(self.number_illum))
else:
illu_order = range(self.number_illum)
for batch_idx in range(num_batch):
if batch_update:
gradient_batch[:,:,:] = 0.0
if self.configs.batch_size == 0:
illu_indices = illu_order
else:
illu_indices = illu_order[batch_idx * self.configs.batch_size : (batch_idx+1) * self.configs.batch_size]
for illu_idx in illu_indices:
#forward scattering
fx_illu = self.fx_illu_list[illu_idx]
fy_illu = self.fy_illu_list[illu_idx]
fields = self._forwardMeasure(fx_illu, fy_illu, obj = obj_gpu)
#calculate error
measurement = af.to_array(measurements[:,:,:,illu_idx].astype(np_complex_datatype))
if self.configs.recon_from_field:
residual = fields["forward_scattered_field"] - measurement
else:
if self.configs.cost_criterion == "intensity":
residual = af.abs(fields["forward_scattered_field"])**2 - measurement**2
elif self.configs.cost_criterion == "amplitude":
residual = af.abs(fields["forward_scattered_field"]) - measurement
cost += af.sum(residual*af.conjg(residual)).real
#calculate gradient
if batch_update:
gradient_batch[:, :, :] += self._computeGradient(fields, measurement)[0]
else:
gradient = self._computeGradient(fields, measurement)
obj_gpu[:, :, :] -= stepsize * gradient
if verbose:
if self.number_illum > 1:
print("gradient update of illumination {:03d}/{:03d}.".format(illu_counter, self.number_illum), end="\r")
illu_counter += 1
fields = None
residual = None
gradient = None
measurement = None
pupil = None
af.device_gc()
if batch_update:
obj_gpu[:, :, :] -= stepsize * gradient_batch
if np.isnan(obj_gpu).sum() > 0:
stepsize *= 0.1
self.configs.time_elapsed = timer.elapsed
print("WARNING: Gradient update diverges! Resetting stepsize to %3.2f" %(stepsize))
t_k = 1.0
continue
# L2 regularizer
obj_gpu[:, :, :] -= stepsize * reg_term * obj_gpu
#record total error
self.configs.error.append(cost + reg_term * af.sum(obj_gpu*af.conjg(obj_gpu)).real)
#Prox operators
af.device_gc()
obj_gpu = self._regularizer_obj.applyRegularizer(obj_gpu)
if flag_FISTA:
#check convergence
if iteration > 0:
if self.configs.error[-1] > self.configs.error[-2]:
if restart:
t_k = 1.0
self._x[:, :, :] = y_k
# stepsize *= 0.8
print("WARNING: FISTA Restart! Error: %5.5f" %(np.log10(self.configs.error[-1])))
if run_callback:
callback(self._x, self.configs)
continue
else:
print("WARNING: Error increased! Error: %5.5f" %(np.log10(self.configs.error[-1])))
#FISTA auxiliary variable
y_k1 = np.array(obj_gpu)
if len(y_k1.shape) < 3:
y_k1 = y_k1[:,:,np.newaxis]
#FISTA update
t_k1 = 0.5*(1.0 + (1.0 + 4.0*t_k**2)**0.5)
beta = (t_k - 1.0) / t_k1
self._x[:, :, :] = y_k1 + beta * (y_k1 - y_k)
t_k = t_k1
y_k = y_k1.copy()
else:
#check convergence
temp = np.array(obj_gpu)
if len(temp.shape) < 3:
temp = temp[:,:,np.newaxis]
self._x[:, :, :] = temp
if iteration > 0:
if self.configs.error[-1] > self.configs.error[-2]:
print("WARNING: Error increased! Error: %5.5f" %(np.log10(self.configs.error[-1])))
stepsize *= 0.8
if verbose:
print("iteration: %d/%d, error: %5.5f, elapsed time: %5.2f seconds" %(iteration+1, max_iter, np.log10(self.configs.error[-1]), timer.elapsed))
if run_callback:
callback(self._x, self.configs)
self.configs.time_elapsed = timer.elapsed
return self._x
af.display(af.hypot(a, b)) af.display(af.sin(a)) af.display(af.cos(a)) af.display(af.tan(a)) af.display(af.asin(a)) af.display(af.acos(a)) af.display(af.atan(a)) af.display(af.atan2(a, b)) c = af.cplx(a) d = af.cplx(a,b) af.display(c) af.display(d) af.display(af.real(d)) af.display(af.imag(d)) af.display(af.conjg(d)) af.display(af.sinh(a)) af.display(af.cosh(a)) af.display(af.tanh(a)) af.display(af.asinh(a)) af.display(af.acosh(a)) af.display(af.atanh(a)) a = af.abs(a) b = af.abs(b) af.display(af.root(a, b)) af.display(af.pow(a, b)) af.display(af.pow2(a)) af.display(af.exp(a))
def conj(self):
if not numpy.issubdtype(self.dtype, numpy.complex):
return afnumpy.copy(self)
s = arrayfire.conjg(self.d_array)
return ndarray(self.shape, dtype=pu.typemap(s.dtype()), af_array=s)
def vdot(a, b):
s = arrayfire.dot(arrayfire.conjg(a.flat.d_array), b.flat.d_array)
return afnumpy.ndarray((), dtype=a.dtype, af_array=s)[()]
