LION.optimizers.AdversarialRegularizerSolver

Classes

ARSolver(model, optimizer[, ...])

Trains an adversarial regulariser (Lunz et al., 2018).

---

class LION.optimizers.AdversarialRegularizerSolver.ARSolver(model, optimizer, operator_params=None, solver_params=None, model_regularization=None, scheduler=None, scheduler_step='epoch', verbose=1, device=None, seed=0, deterministic=False)

Bases: LIONsolver

Trains an adversarial regulariser (Lunz et al., 2018).

Training compares clean images with reconstructions of the measurements (recon_fn, by default the operator’s: FDK for CT) (WGAN loss with gradient penalty). The WGAN loss doesn’t measure reconstructions, so validation and testing reconstruct with a LearnedRegularizerReconstructor you make and give (set_validation(..., reconstructor=...)): the solver uses it with the regulariser being trained. Its settings (e.g. alpha, which estimate_alpha estimates as in the paper) are the reconstructor’s. To reconstruct with a trained regulariser, use the same class with the model.

Every solver takes (model, optimizer, loss_fn, operator_params, solver_params, model_regularization, scheduler, scheduler_step, verbose, device, seed, deterministic), in that order (ARSolver has no loss_fn).

operator_params: the parameters of the operator (for CT, the geometry); the model’s, if it has them.

solver_params: the solver’s settings, from default_parameters().

model_regularization: None, or a term added to the training loss of every mini-batch, model_regularization(model, step): a penalty on the model’s weights (step unused) or on what it does with the mini-batch (e.g. LION.losses.jacobian_penalty). step has input (what the model was given), output, target (what the output is compared with; None for solvers without one) and forward (the model as the solver calls it, e.g. with the noise level). Recorded as loss_fn is.

scheduler: a torch learning-rate scheduler made on optimizer (e.g. torch.optim.lr_scheduler.StepLR(optimizer, 30)), or None. It steps after every epoch (scheduler_step="epoch", torch’s convention) or every mini-batch ("batch", e.g. for OneCycleLR). ReduceLROnPlateau steps with the validation value of the epochs that validate, or with the training loss when there is no validation. Its state is saved in checkpoints, and the training record has it and the learning rate of every epoch.

seed: the training seed. When training starts, torch (CPU and CUDA), numpy and Python’s random are seeded with it, so the same solver, model and data give the same training, and checkpoints keep the random state, so a resumed run continues the same way. The randomness within a run (shuffling, noise a solver adds) still changes every batch and epoch. Two runs with the same seed give the same model: use different seeds for an ensemble. None: unseeded.

deterministic: use only deterministic algorithms (torch’s use_deterministic_algorithms, deterministic cuDNN), so that on a GPU two runs with the same seed give exactly the same model, and a resumed run the uninterrupted one. Slower; a few operations then raise (torch names them). It applies to the whole program from here on. Off, GPU runs with the same seed differ slightly (e.g. ~1e-3 in the weights after a few epochs). On a GPU it needs CUBLAS_WORKSPACE_CONFIG=:4096:8 before cuBLAS starts: it is set here if it isn’t (exporting it before the script is surer).

verbose: how much progress is shown. 0: nothing; 1 (default): one progress bar for training (every mini-batch of every epoch, with the epoch, the loss, the last validation value and the time left) and one for testing; 2: also the validation bars and messages (each epoch’s losses, checkpoints, the seed). True is 1, False 0.

reconstructed_with = 'LearnedRegularizerReconstructor'

static default_parameters()

lambd_gp: the weight of the gradient penalty in the WGAN training loss (0.1). recon_fn: how measurements are reconstructed into the “bad” reconstructions the regulariser learns to tell from clean images, recon_fn(y, op); “default”: the operator’s default reconstruction (FDK for CT), replaced by that function when the solver is made.

set_validation(validation_loader, validation_freq, validation_fn=None, validation_fname=None, reconstructor=None)

As LIONsolver.set_validation; reconstructor (a LearnedRegularizerReconstructor) is needed. By default, validate with the MSE between reconstruction and target.

set_testing(test_loader, testing_fn=None, reconstructor=None)

As LIONsolver.set_testing; reconstructor is needed.

wgan_loss(y, target_batch)

The WGAN loss with gradient penalty: the regulariser’s mean on clean images (target_batch) minus its mean on reconstructions of y, plus lambd_gp times the gradient penalty at random points between them.

mini_batch_step(y, target_batch)

The WGAN loss of the batch (wgan_loss).

static cite(cite_format='bib')

Where ARSolver comes from: its citation, BibTeX (“bib”) or “MLA”.