sylo#

Initialisations for sylo-based neural networks.

sylo_init#

hypercoil.init.sylo.sylo_init(*, shape: Tuple[int, ...], shape_R: Tuple[int, ...] | None = None, negative_slope: float = 0, mode: Literal['fan_in', 'fan_out'] = 'fan_in', init_distr: Literal['uniform', 'normal'] = 'uniform', nonlinearity: str = 'leaky_relu', psd: bool = False, key: PRNGKey) → Tensor | Tuple[Tensor, Tensor][source]#

SyloInitialiser#

class hypercoil.init.sylo.SyloInitialiser(negative_slope: 'float' = 0, mode: "Literal['fan_in', 'fan_out']" = 'fan_in', init_distr: "Literal['uniform', 'normal']" = 'uniform', nonlinearity: 'str' = 'leaky_relu', psd: 'bool' = False, mapper: 'Optional[Type[MappedParameter]]' = None)[source]#

Methods

__call__(model, *[, where])

Call self as a function.

init(model, *[, mapper, negative_slope, ...])

Initialise a parameter using the specified initialiser and mapper.