NectarGAN API - Scheduling
The NectarGAN API provides a modular, highly configurable solution to managing scheduling for any parameter in your models.
Scheduling dataclass
The core of the NectarGAN scheduling system is the Schedule dataclass, an easy to use tool which (along with the scheduling function system) allows you to quickly define complex schedules which can be applied to any paramaeter in your model with relative ease.
Please see here for more information.
Schedule Functions
The NectarGAN scheduling system works via a drop-in schedule function system. This system allows you to define reusable schedule functions, and easily drop them in to anything you'd like to schedule.
Please see here for more information.
Scheduler Classes
Scheduling in the NectarGAN API is managed by two main classes: a general scheduler, which can be used for basically anything, and a compatibility wrapper around PyTorch's native torch.optim.lr_scheduler, allowing you to use scheduling functions interchangeably while retaining the native scheduler's deep integration with the native optimizers.
Please see here for more information.
Scheduler Integrations
Schedules are deeply integrated into other core components of the NectarGAN API.
Please see here for more information.