dynax.util

Functions

broadcast_right(arr, target)	Append trailing size-1 axes to arr until it has the same ndim as target.
dim2shape(x)	Convert a dimension specifier to a shape tuple.
monkeypatch_pretty_print()
mse(target, prediction[, axis])	Compute mean-squared error.
multitone(length, num_tones[, first_tone])	Create a periodic bandpassed multitone signal.
nmse(target, prediction[, axis])	Compute normalized mean-squared error.
nrmse(target, prediction[, axis])	Compute normalized root mean-squared error.
pretty(tree)	Return a pretty-formatted string representation of an equinox pytree.
value_and_jacfwd(fun, x)	Evaluate fun and its forward-mode Jacobian at x.
value_and_jacrev(fun, x)	Evaluate fun and its reverse-mode Jacobian at x.

dynax.util.value_and_jacfwd(fun, x)

Evaluate fun and its forward-mode Jacobian at x.

Parameters:

• fun (Callable) – Function whose Jacobian is to be computed.

• x (Array) – Point at which function and Jacobian is evaluated.

Returns:

tuple[Array, Array] – Tuple (y, jac) of the function value and Jacobian matrix at x.

From this issue.

dynax.util.value_and_jacrev(fun, x)

Evaluate fun and its reverse-mode Jacobian at x.

Parameters:

• fun (Callable) – Function whose Jacobian is to be computed.

• x (Array) – Point at which function and Jacobian is evaluated.

Returns:

tuple[Array, Array] – Tuple (y, jac) of the function value and Jacobian matrix at x.

From this issue.

dynax.util.mse(target, prediction, axis=0)

Compute mean-squared error.

Return type:

Array

dynax.util.nmse(target, prediction, axis=0)

Compute normalized mean-squared error.

Return type:

Array

dynax.util.nrmse(target, prediction, axis=0)

Compute normalized root mean-squared error.

Return type:

Array

dynax.util.pretty(tree)

Return a pretty-formatted string representation of an equinox pytree.

dynax.util.broadcast_right(arr, target)

Append trailing size-1 axes to arr until it has the same ndim as target.

dynax.util.dim2shape(x)

Convert a dimension specifier to a shape tuple.

Parameters:

x (int | Literal['scalar']) – Either “scalar” or a non-negative integer number of dimensions.

Returns:

tuple – () if x is “scalar”, otherwise (x,).

dynax.util.multitone(length, num_tones, first_tone=0)

Create a periodic bandpassed multitone signal.

Has flat spectrum and crest factor <= 2.

Return type:

Array

From: S. Boyd, “Multitone Signals with Low Crest Factor”

IEEE Transactions on Circuits and Systems, CAS-33(10):1018-1022, October 1986