auto_uncertainties.jittable_function_wrapper

Attributes

P
R

Functions

elementwise_value_and_grad(g)
propagate_uncertainties(func[, implement_mode])	A decorator to propagate uncertainties through a given function using

Module Contents

auto_uncertainties.jittable_function_wrapper.P

auto_uncertainties.jittable_function_wrapper.R

auto_uncertainties.jittable_function_wrapper.elementwise_value_and_grad(g)

auto_uncertainties.jittable_function_wrapper.propagate_uncertainties(func, implement_mode='same_shape')

A decorator to propagate uncertainties through a given function using first-order Taylor expansion.

Parameters:

• func (callable) – The function through which to propagate uncertainties. This function should be compatible with JAX for automatic differentiation.

• implement_mode (str)

Returns:

A new function that propagates uncertainties when called with Uncertainty objects.

Return type:

callable

Notes

This decorator uses JAX’s automatic differentiation to compute gradients, which are then used to propagate uncertainties. The function being decorated should be compatible with JAX, meaning it should use JAX-compatible operations.

Examples

>>> import numpy as np
>>> import jax.numpy as jnp
>>> from auto_uncertainties import Uncertainty, propagate_uncertainties
>>> @propagate_uncertainties
... def my_function(x, y):
...     return x * jnp.sin(y)
>>> x = Uncertainty(2.0, 0.1)  # Value 2.0 with uncertainty 0.1
>>> y = Uncertainty(np.pi / 4, 0.01)  # Value pi/4 with uncertainty 0.01
>>> result = my_function(x, y)