Files
cdtools/CDTools/models/__init__.py
T

141 lines
4.6 KiB
Python

from __future__ import division, print_function, absolute_import
import torch as t
from torch.utils import data as torchdata
#
# This is unrelated, but it will then be important to be able to save and load
# models easily from a predefined format. To be honest, this could just
# literally be by pickling the model. They could also be saved out as
# state_dicts or via torch.save. I think it's best to just save the whole
# model - I lose out on the modularity of just saving the state_dict, but
# I gain in it being easy to reload the non-learned aspects of the model,
# like the wavelength and sample geometry. Remember that it's important
# that the final outputs of the reconstructions are transferrable to other
# places
#
#
# For now, just save/load model via the built-in t.save() and t.load()
# functions
#
class CDIModel(t.nn.Module):
"""This base model defines all the functions that must be exposed for a valid CDIModel subclass
Most of the functions only raise a NotImplementedError at this level and
must be explicitly defined by any subclass. The functions required can be
split into several subsections:
Creation:
from_dataset : a function to create a CDIModel from an appropriate CDataset
Simulation:
interaction : a function to simulate exit waves from experimental parameters
forward_propagator : the propagator from the experiment plane to the detector plane
backward_propagator : the propagator from the detector plane to the experiment plane
measurement : a function to simulate the detector readout from a detector plane wavefront
forward : predefined, the entire stacked forward model
loss : the loss function to report and use for automatic differentiation
simulation : predefined, simulates a stack of detector images from the forward model
simulate_to_dataset : a function to create a CDataset from the simulation defined in the model
Reconstruction:
AD_optimize : predefined, a generic automatic differentiation reconstruction
Adam_optimize : predefined, sensible automatic differentiation reconstruction using ADAM
The work of defining the various subclasses boils down to creating an
appropriate implementation for this set of functions.
"""
def from_dataset(self, dataset):
raise NotImplementedError()
def interaction(self, *args):
raise NotImplementedError()
def forward_propagator(self, exit_wave):
raise NotImplementedError()
def backward_propagator(self, detector_wave):
raise NotImplementedError()
def measurement(self, detector_wave):
raise NotImplementedError()
def forward(self, *args):
return self.measurement(self.forward_propagator(self.interaction(*args)))
def loss(self, sim_data, real_data):
raise NotImplementedError()
# I know this is silly but it makes it clear this should be explicitly
# overwritten
def to(self, *args, **kwargs):
super(CDIModel,self).to(*args,**kwargs)
def simulate(self, args_list):
return t.Tensor([self.forward(*args) for args in args_list])
def simulate_to_dataset(self, args_list):
raise NotImplementedError()
def AD_optimize(self, iterations, data_loader, optimizer, scheduler=None):
for it in range(iterations):
loss = 0
N = 0
for inputs, patterns in data_loader:
N += patterns.shape[0]
def closure():
optimizer.zero_grad()
sim_patterns = self.forward(*inputs)
if hasattr(self, 'mask'):
loss = self.loss(patterns,sim_patterns, mask=self.mask)
else:
loss = self.loss(patterns,sim_patterns)
loss.backward()
return loss
loss += optimizer.step(closure).detach().cpu().numpy()
loss /= N
if scheduler is not None:
scheduler.step(loss)
yield loss
def Adam_optimize(self, iterations, dataset, batch_size=15, lr=0.005):
# Make a dataloader
data_loader = torchdata.DataLoader(dataset, batch_size=batch_size,
shuffle=True)
# Define the optimizer
optimizer = t.optim.Adam(self.parameters(), lr = lr)
return self.AD_optimize(iterations, data_loader, optimizer)
from CDTools.models.simple_ptycho import SimplePtycho