ml.lr_schedulers.gan

GAN learning rate scheduler wrapper.

This wrapper allows for downstream users to set different learning rate schedules for the generator and discriminator of a GAN.

This class is used by the GAN trainer interface and shouldn’t be used elsewhere.

class ml.lr_schedulers.gan.GenerativeAdversarialNetworkLRSchedulerConfig(name: str = '???', generator: Any = '???', discriminator: Any = '???')

Bases: BaseLRSchedulerConfig

generator: Any = '???'

discriminator: Any = '???'

classmethod update(config: DictConfig) → DictConfig

Runs post-construction config update.

Parameters:

config – The config to update

classmethod resolve(config: GenerativeAdversarialNetworkLRSchedulerConfig) → None

Runs post-construction config resolution.

Parameters:

config – The config to resolve

class ml.lr_schedulers.gan.GenerativeAdversarialNetworkLRScheduler(config: GenerativeAdversarialNetworkLRSchedulerConfig)

Bases: BaseLRScheduler[GenerativeAdversarialNetworkLRSchedulerConfig]

get_lr_scale(state: State) → float

Given a state, returns the current learning rate.

Parameters:

state – The current trainer state

Returns:

The computed learning rate to use