smoothness: Backwards differences

smoothness

hypercoil.loss.smoothness(X: Tensor, *, n: int = 1, pad_value: float | None = None, axis: int = -1, key: PRNGKey | None = None) → Tensor

Smoothness score function.

This loss penalises large or sudden changes in the input tensor. It is currently a thin wrapper around jax.numpy.diff.

Warning

This function returns both positive and negative values, and so should probably not be used with a scalarisation map like mean_scalarise or sum_scalarise. Instead, maps like meansq_scalarise or vnorm_scalarise with either the p=1 or p=inf options might be more appropriate.

Parameters:

XTensor

Input tensor.

n: int, optional (default: 1)

Number of times to differentiate using the backwards differences method.

axisint, optional (default: -1)

Axis defining the slice of the input tensor over which differences are computed

pad_valuefloat, optional (default: None)

Arguments to jnp.diff. Values to prepend to the input along the specified axis before computing the difference.

SmoothnessLoss

class hypercoil.loss.SmoothnessLoss(nu: float = 1.0, name: str | None = None, *, n: int = 1, pad_value: float | None = None, axis: int = -1, scalarisation: Callable | None = None, key: 'jax.random.PRNGKey' | None = None)

Smoothness loss function.

This loss penalises large or sudden changes in the input tensor. It is currently a thin wrapper around jax.numpy.diff.

Warning

This function returns both positive and negative values, and so should probably not be used with a scalarisation map like mean_scalarise or sum_scalarise. Instead, maps like meansq_scalarise or vnorm_scalarise with either the p=1 or p=inf options might be more appropriate.

Parameters:

name: str

Designated name of the loss function. It is not required that this be specified, but it is recommended to ensure that the loss function can be identified in the context of a reporting utilities. If not explicitly specified, the name will be inferred from the class name and the name of the scoring function.

nu: float

Loss strength multiplier. This is a scalar multiplier that is applied to the loss value before it is returned. This can be used to modulate the relative contributions of different loss functions to the overall loss value. It can also be used to implement a schedule for the loss function, by dynamically adjusting the multiplier over the course of training.

n: int, optional (default: 1)

Number of times to differentiate using the backwards differences method.

axisint, optional (default: -1)

Axis defining the slice of the input tensor over which differences are computed

pad_valuefloat, optional (default: None)

Arguments to jnp.diff. Values to prepend to the input along the specified axis before computing the difference.

scalarisation: Callable

The scalarisation function to be used to aggregate the values returned by the scoring function. This function should take a single argument, which is a tensor of arbitrary shape, and return a single scalar value. By default, the L1 norm scalarisation is used.

Methods

__call__(X, *[, key])

Call self as a function.