abel package

Submodules

abel.abel module

class abel.abel.ABEL(optimizer, decay: float = 0.1, last_epoch: int = - 1, current_norm: Optional[torch.Tensor] = None, norm_t_1: Optional[torch.Tensor] = None, norm_t_2: Optional[torch.Tensor] = None, verbose: bool = False)

Bases: torch.optim.lr_scheduler._LRScheduler

Automatic, Bouncing into Equilibration Learning rate scheduler.

Parameters

• optimizer (torch.optim.Optimizer) – torch based optimizer

• decay (float) – LR decay(default=0.1)

• last_epoch (int) – Last executed epoch(default=-1)

• current_norm (torch.Tensor) – current weight norm of model(default=None)

• norm_t_1 (torch.Tensor) – t-1 weight norm of model(default=None)

• norm_t_2 (torch.Tensor) – t-2 weight norm of model(default=None)

• verbose (bool) – Verbosity(default=False)

abel.utils module

abel.utils.get_weight_norm(param_groups: Iterable) → torch.Tensor

Returns weight norm of the param groups

Parameters

param_groups (Iterable) – List of parameters of the model