def getFlopsPerIter(self) -> int:
if self.n_validation > 0:
e = RuntimeWarning("n_validation > 0 can give misleading flops.")
logger.warning(e)
if hasattr(self, "session"):
e = RuntimeWarning(
"Calculating computational cost after previously initializing the static graph can "
+
"include inessential calculations (like training and/or validation loss value) "
+ "and thus give misleading results.")
logger.warning(e)
else:
with self.graph.as_default():
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
logger.info("Initializing the session.")
self.session = tf.Session(config=config)
self.session.run(tf.global_variables_initializer())
for opt in self.optimizers:
self.session.run(opt.minimize_op)
total_flops = getComputationalCostInFlops(self.graph)
return total_flops
def attachTensorflowOptimizerForVariable(
self,
variable_name: str,
optimizer_type: str,
optimizer_init_args: dict = None,
optimizer_minimize_args: dict = None,
initial_update_delay: int = 0,
update_frequency: int = 1,
checkpoint_frequency: int = 100):
"""Attach an optimizer for the specified variable.
Parameters
----------
variable_name : str
Name (string) associated with the chosen variable (to be optimized) in the forward model.
optimizer_type : str
Name of the standard optimizer chosen from availabe options in options.Options.
optimizer_init_args : dict
Dictionary containing the key-value pairs required for the initialization of the desired optimizer.
optimizer_minimize_args : dict
Dictionary containing the key-value pairs required to define the minimize operation for the desired
optimizer.
initial_update_delay : int
Number of iterations to wait before the minimizer is first applied. Defaults to 0.
update_frequency : int
Number of iterations in between minimization calls. Defaults to 1.
checkpoint_frequency : int
Number of iterations between creation of checkpoints of the optimizer. Not implemented.
"""
optimization_all = ptychoSampling.reconstruction.options.OPTIONS[
"tf_optimization_methods"]
self._checkConfigProperty(optimization_all, optimizer_type)
self._checkAttr("_train_loss_t", "optimizer")
if variable_name not in self.fwd_model.model_vars:
e = ValueError(
f"{variable_name} is not a supported variable in {self.fwd_model}"
)
logger.error(e)
raise e
var = self.fwd_model.model_vars[variable_name]["variable"]
if optimizer_minimize_args is None:
optimizer_minimize_args = {}
elif ("loss"
in optimizer_minimize_args) or ("var_list"
in optimizer_minimize_args):
warning = (
"Target loss and optimization variable are assigned by default. "
+
"If custom processing is desired, use _attachCustomOptimizerForVariable directly."
)
logger.warning(warning)
optimizer_minimize_args["loss"] = self._train_loss_t
optimizer_minimize_args["var_list"] = [var]
self._attachCustomOptimizerForVariable(
optimization_all[optimizer_type], optimizer_init_args,
optimizer_minimize_args, initial_update_delay, update_frequency)
def createObj(obj_params_dict: dict,
det_3d_shape: Tuple[int, int, int],
obj_pixel_size: Tuple[float, float, float],
upsampling_factor: float = 1.0) -> Obj:
logger.info("Creating new crystal cell.")
while True:
obj_crystal = Simulated3DCrystalCell(**obj_params_dict)
if det_3d_shape[
1] * upsampling_factor >= obj_crystal.bordered_array.shape[
1]:
break
else:
logger.warning(
"Generated object width is larger than detector x-width. Trying again."
)
logger.info(
'Adding x and z borders to crystal cell based on detector parameters.'
)
# Ensuring that the number of pixels in the x dimension matches that in the detector.
padx = (det_3d_shape[1] - obj_crystal.bordered_array.shape[1]) // 2
# Calculating the padding needed to accomodate all feasible probe translations (w probe-obj interaction)
padz, _ = Simulation.calculateObjBorderAfterRotation(
0, det_3d_shape[2], obj_crystal.bordered_array.shape[1], 0)
pad_shape = (0, padx, padz)
obj = CustomObjFromArray(array=obj_crystal.bordered_array,
border_shape=np.vstack(
(pad_shape, pad_shape)).T,
border_const=0,
pixel_size=obj_pixel_size)
return obj
def _checkFinalized(self):
if not hasattr(self, "dataframe"):
e = AttributeError(
"Cannot add item to the log file after starting the optimization. "
+
"The log file remains unchanged. Only the print output is affected."
)
logger.warning(e)
def _calculateWavefront(self) -> None:
"""Calculating the airy pattern then propagating it by `defocus_dist`."""
if self.width_dist[0] > 0:
logger.warning(
"Warning: if width at the focus is supplied, " +
"any supplied focal_length and aperture radius are ignored.")
self.aperture_radius = None
self.focal_length = None
# Assuming resolution equal to pixel pitch and focal length of 10.0 m.
focal_length = 10.0
focus_radius = self.width_dist[0] / 2
# note that jinc(1,1) = 1.22 * np.pi
aperture_radius = (self.wavelength * 1.22 * np.pi * focal_length /
2 / np.pi / focus_radius)
else:
if None in [self.aperture_radius, self.focal_length]:
e = ValueError(
'Either focus_radius_npix or BOTH aperture_radius and focal_length must be supplied.'
)
logger.error(e)
raise e
aperture_radius = self.aperture_radius
focal_length = self.focal_length
npix_oversampled = max(
self.oversampling_npix,
self.shape[-1]) if self.oversampling else self.shape[0]
pixel_pitch_aperture = self.wavelength * focal_length / (
npix_oversampled * self.pixel_size[0])
x = np.arange(-npix_oversampled // 2,
npix_oversampled // 2) * pixel_pitch_aperture
r = np.sqrt(x**2 + x[:, np.newaxis]**2).astype('float32')
circ_wavefront = np.zeros(r.shape)
circ_wavefront[r < aperture_radius] = 1.0
circ_wavefront[r == aperture_radius] = 0.5
probe_vals = np.fft.fftshift(
np.fft.fft2(np.fft.fftshift(circ_wavefront), norm='ortho'))
n1 = npix_oversampled // 2 - self.shape[0] // 2
n2 = npix_oversampled // 2 + self.shape[1] // 2
scaling_factor = np.sqrt(self.n_photons /
np.sum(np.abs(probe_vals)**2))
wavefront_array = probe_vals[n1:n2, n1:n2].astype(
'complex64') * scaling_factor
wavefront_array = np.fft.fftshift(wavefront_array)
#self.wavefront = self.wavefront.update(array=wavefront_array)
self.wavefront = Wavefront(wavefront_array,
wavelength=self.wavelength,
pixel_size=self.pixel_size)
self._propagateWavefront()
def getFlopsPerIter(self) -> dict:
if self.n_validation > 0:
e = RuntimeWarning("n_validation > 0 can give misleading flops.")
logger.warning(e)
if hasattr(self, "session"):
e = RuntimeWarning(
"Calculating computational cost after previously initializing the static graph can "
+
"include inessential calculations (like training and/or validation loss value) "
+ "and thus give misleading results.")
logger.warning(e)
else:
self.finalizeSetup()
self.session.run(self._new_train_batch_op)
for opt in self.optimizers:
self.session.run(opt.minimize_op)
total_flops = getComputationalCostInFlops(self.graph)
cg_init_flops = getComputationalCostInFlops(
self.graph,
keywords=[("opt_minimize_step", "conjugate_gradient", "cg_init")],
# "obj_opt_minimize_step/while/conjugate_gradient/cg_while"],
exclude_keywords=False)
cg_while_flops = getComputationalCostInFlops(
self.graph,
keywords=[("opt_minimize_step", "conjugate_gradient", "cg_while")],
# "obj_opt_minimize_step/while/conjugate_gradient/cg_while"],
exclude_keywords=False)
proj_ls_flops = getComputationalCostInFlops(self.graph,
keywords=[
("opt_minimize_step",
"proj_ls_linesearch")
],
exclude_keywords=False)
opt_only_flops = getComputationalCostInFlops(self.graph,
keywords=["opt"],
exclude_keywords=False)
flops_without_cg_ls = total_flops - cg_while_flops - proj_ls_flops
d = {
"total_flops": total_flops,
"obj_cg_flops": cg_while_flops,
"probe_cg_flops": 0,
"obj_proj_ls_flops": proj_ls_flops,
"probe_proj_ls_flops": 0,
"obj_only_flops": opt_only_flops,
"probe_only_flops": 0,
"flops_without_cg_ls": flops_without_cg_ls
}
return d
def getFlopsPerIter(self) -> dict:
if self.n_validation > 0:
e = RuntimeWarning("n_validation > 0 can give misleading flops.")
logger.warning(e)
if hasattr(self, "session"):
e = RuntimeWarning(
"Calculating computational cost after previously initializing the static graph can "
+
"include inessential calculations (like training and/or validation loss value) "
+ "and thus give misleading results.")
logger.warning(e)
else:
with self.graph.as_default():
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
logger.info("Initializing the session.")
self.session = tf.Session(config=config)
self.session.run(tf.global_variables_initializer())
for opt in self.optimizers:
self.session.run(opt.minimize_op)
total_flops = getComputationalCostInFlops(self.graph)
ls_flops = getComputationalCostInFlops(self.graph,
keywords=[
("opt_minimize_step",
"backtracking_linesearch")
],
exclude_keywords=False)
opt_only_flops = getComputationalCostInFlops(self.graph,
keywords=["opt"],
exclude_keywords=False)
flops_without_ls = total_flops - ls_flops
d = {
"total_flops": total_flops,
"obj_ls_flops": ls_flops,
"probe_ls_flops": 0,
"obj_only_flops": opt_only_flops,
"probe_only_flops": 0,
"flops_without_ls": flops_without_ls
}
return d
def __init__(self,
wavelength: float,
pixel_size: Tuple[float, ...],
shape: Tuple[int, ...],
n_photons: float,
defocus_dist: float = 0,
center_dist: Tuple[float, float] = (0, 0),
width_dist: Tuple[float, float] = (0, 0),
center_npix: Tuple[int, int] = None,
width_npix: Tuple[int, int] = None,
check_propagation_with_gaussian_fit: bool = False,
apodize: bool = False) -> None:
self.wavelength = wavelength
self.shape = shape
self.pixel_size = pixel_size
self.n_photons = n_photons
self.defocus_dist = defocus_dist
self.center_dist = center_dist
self.width_dist = width_dist
self.apodize = apodize
if center_npix is not None:
logger.warning(
'If center_npix is supplied, then any supplied center_dist is ignored.'
)
self.center_npix = center_npix
self.center_dist = np.array(center_npix) * np.array(
self.pixel_size)
if width_npix is not None:
logger.warning(
'If width_npix is supplied, then any supplied width_dist is ignored.'
)
self.width_npix = width_npix
self.width_dist = np.array(width_npix) * np.array(self.pixel_size)
#wavefront_array = np.zeros((npix, npix), dtype='complex64')
#self.wavefront = propagators.Wavefront(wavefront_array,
self.wavefront = Wavefront(np.zeros(shape),
wavelength=wavelength,
pixel_size=pixel_size)
self.photons_flux = n_photons / (shape[-1] * shape[-2])
self.check_propagation_with_gaussian_fit = check_propagation_with_gaussian_fit
def __init__(self,
wavelength: float = 1.5e-10,
obj: Obj = None,
obj_params: dict = None,
probe: Probe = None,
probe_params: dict = None,
scan_grid: ScanGrid = None,
scan_grid_params: dict = None,
detector: Detector = None,
detector_params: dict = None,
poisson_noise: bool = True,
random_shift_pix: Tuple[int, int] = None,
upsampling_factor: int = 1,
background_scaling: float=1e-8,
background_constant: float=0) -> None:
self.wavelength = wavelength
self.upsampling_factor = upsampling_factor
self.poisson_noise = poisson_noise
self.background_scaling = background_scaling
self.background_constant = background_constant
if obj or probe or scan_grid or detector:
logger.warning("If one (or all) of obj, probe, scan_grid, or detector is supplied, "
+ "then the corresponding _params parameter is ignored.")
if detector is not None:
self.detector = detector
self._detector_params = {}
else:
detector_params = {} if detector_params is None else detector_params
self._detector_params = DetectorParams(**detector_params)
self.detector = Detector(**dt.asdict(self._detector_params))
detector_support_size = np.asarray(self.detector.pixel_size) * self.detector.shape
obj_pixel_size = self.wavelength * self.detector.obj_dist / (detector_support_size * self.upsampling_factor)
probe_shape = np.array(self.detector.shape) * self.upsampling_factor
if obj is not None:
if obj.pixel_size is not None and np.any(obj.pixel_size != obj_pixel_size):
e = ValueError("Mismatch between the provided pixel size and the pixel size calculated from scan "
+ "parameters.")
logger.error(e)
raise e
obj.pixel_size = obj_pixel_size
else:
obj_params = {} if obj_params is None else obj_params
self._obj_params = ObjParams(**obj_params)
self.obj = Simulated2DObj(**dt.asdict(self._obj_params))
if probe is not None:
check = (np.any(probe.shape != probe_shape)
or (probe.wavelength != self.wavelength)
or np.any(probe.pixel_size != obj_pixel_size))
if check:
e = ValueError("Supplied probe parameters do not match with supplied scan and detector parameters.")
logger.error(e)
raise e
self.probe = probe
else:
probe_params = {} if probe_params is None else probe_params
self._probe_params = ProbeParams(**probe_params)
self.probe = FocusCircularProbe(wavelength=wavelength,
pixel_size=obj_pixel_size,
shape=probe_shape,
**dt.asdict(self._probe_params))
if scan_grid is not None:
self.scan_grid = scan_grid
self._scan_grid_params = None
else:
scan_grid_params = {} if scan_grid_params is None else scan_grid_params
self._scan_grid_params = GridParams(**scan_grid_params)
self.scan_grid = RectangleGrid(obj_w_border_shape=self.obj.bordered_array.shape,
probe_shape=self.probe.shape,
obj_pixel_size=obj_pixel_size,
**dt.asdict(self._scan_grid_params))
#if self._scan_grid_params.random_shift_dist is not None:
# raise
if random_shift_pix is not None:
self.scan_grid = self._updateGridWithRandomPixShifts(self.scan_grid, random_shift_pix)
#elif self._scan_grid_params.random_shift_ratio is not None:
# raise
self.scan_grid.checkOverlap()
self._calculateDiffractionPatterns()
def __init__(self,
wavelength: float = 1.5e-10,
obj: Obj = None,
probe_3d: Probe3D = None,
scan_grid: BraggPtychoGrid = None,
detector: Detector = None,
poisson_noise: bool = True,
upsampling_factor: int = 1) -> None:
self.wavelength = wavelength
self.upsampling_factor = upsampling_factor
self.poisson_noise = poisson_noise
two_theta = scan_grid.two_theta if scan_grid is not None else AngleGridParams(
).two_theta
if detector is not None:
logger.info("Using supplied detector info.")
self.detector = detector
else:
logger.info("Creating new detector.")
self.detector = Detector(**dt.asdict(DetectorParams()))
det_3d_shape = (1, *self.detector.shape)
obj_xz_nyquist_support = self.wavelength * self.detector.obj_dist / np.array(
self.detector.pixel_size)
obj_xz_pixel_size = obj_xz_nyquist_support / (
np.array(self.detector.shape) * self.upsampling_factor)
# The y pixel size is very ad-hoc
obj_y_pixel_size = obj_xz_nyquist_support[1] / self.detector.shape[1]
obj_pixel_size = (obj_y_pixel_size, *obj_xz_pixel_size)
if obj is not None:
logger.info("Using supplied object.")
self.obj = obj
else:
self.obj = self.createObj(dt.asdict(ObjParams()), det_3d_shape,
obj_pixel_size, upsampling_factor)
probe_xz_shape = np.array(self.detector.shape) * self.upsampling_factor
if probe_3d is not None:
logger.info("Using supplied 3d probe values.")
self.probe_3d = probe_3d
# Note that the probe y shape cannot be determined from the other supplied parameters (I think).
if (np.any(probe_3d.shape[1:] != probe_xz_shape)
or (probe_3d.wavelength != self.wavelength)
or np.any(probe_3d.pixel_size != obj_pixel_size)):
e = ValueError(
f"Mismatch between the supplied probe and the supplied scan parameters."
)
logger.error(e)
raise e
else:
self.probe_3d = self.createProbe3D(dt.asdict(Probe2DParams()),
wavelength,
np.pi / 2 - two_theta,
probe_xz_shape, obj_pixel_size)
rotate_angle = np.pi / 2 - two_theta
probe_y_pix_before_rotation = self.probe_3d.shape[0] * np.cos(
rotate_angle) // 1
pady, bordered_obj_ypix = self.calculateObjBorderAfterRotation(
rotate_angle, probe_y_pix_before_rotation, self.obj.shape[0],
self.obj.shape[1])
if self.obj.bordered_array.shape[0] < bordered_obj_ypix:
logger.warning(
"Adding zero padding to the object in the y-direction so that the overall object y-width "
+ "covers the entirety of the feasible probe positions.")
self.obj.border_shape = ((pady, pady), self.obj.border_shape[1],
self.obj.border_shape[2])
if scan_grid is not None:
logger.info("Using supplied scan grid.")
self.scan_grid = scan_grid
else:
elems_yz = lambda tup: np.array(tup)[[0, 2]]
scan_grid_params_dict = dt.asdict(
Scan2DGridParams(obj_pixel_size=elems_yz(self.obj.pixel_size)))
self.scan_grid = self.createScanGrid(
scan_grid_params_dict, dt.asdict(AngleGridParams()),
elems_yz(self.obj.bordered_array.shape),
elems_yz(self.probe_3d.shape))
self._calculateDiffractionPatterns()
def __init__(self,
wavelength: float = 1.5e-10,
obj: Obj = None,
obj_params: dict = {},
probe: Probe = None,
probe_params: dict = {},
scan_grid: ScanGrid = None,
scan_grid_params: dict = {},
detector: Detector = None,
detector_params: dict = {},
poisson_noise: bool = True,
upsampling_factor: int = 1) -> None:
self.wavelength = wavelength
self.upsampling_factor = upsampling_factor
self.poisson_noise = poisson_noise
if obj or probe or scan_grid or detector:
logger.warning(
"If one (or all) of obj, probe, scan_grid, or detector is supplied, "
+ "then the corresponding _params parameter is ignored.")
if detector is not None:
self.detector = detector
self._detector_params = {}
else:
self._detector_params = DetectorParams(**detector_params)
self.detector = Detector(**dt.asdict(self._detector_params))
obj_pixel_size = np.array(
self.detector.pixel_size) * self.upsampling_factor
probe_shape = np.array(self.detector.shape) * self.upsampling_factor
if obj is not None:
if obj.pixel_size is not None and np.any(
obj.pixel_size != obj_pixel_size):
e = ValueError(
"Mismatch between the provided pixel size and the pixel size calculated from scan "
+ "parameters.")
logger.error(e)
raise e
obj.pixel_size = obj_pixel_size
else:
self._obj_params = ObjParams(**obj_params)
self.obj = Simulated2DObj(**dt.asdict(self._obj_params))
if probe is not None:
check = (np.any(probe.shape != probe_shape)
or (probe.wavelength != self.wavelength)
or np.any(probe.pixel_size != obj_pixel_size))
if check:
e = ValueError(
"Supplied probe parameters do not match with supplied scan and detector parameters."
)
logger.error(e)
raise e
self.probe = probe
else:
self._probe_params = ProbeParams(**probe_params)
self.probe = GaussianSpeckledProbe(wavelength=wavelength,
pixel_size=obj_pixel_size,
shape=probe_shape,
**dt.asdict(self._probe_params))
if scan_grid is not None:
self.scan_grid = scan_grid
self._scan_grid_params = None
else:
self._scan_grid_params = GridParams(**scan_grid_params)
self.scan_grid = RectangleGrid(
obj_w_border_shape=self.obj.bordered_array.shape,
probe_shape=self.probe.shape,
obj_pixel_size=obj_pixel_size,
**dt.asdict(self._scan_grid_params))
self.scan_grid.checkOverlap()
self._calculateDiffractionPatterns()
