jacobian_from_function#
- cosmicjellyfish.analysis.fisher_operations.jacobian_from_function(function, fiducial, method='fd', step=1e-06)[source]#
Jacobian of a vector function at a fiducial point.
The returned matrix is \(J_{i\alpha} = \partial f_i / \partial x_\alpha\), with rows indexing the outputs of
functionand columns indexing its inputs.- Parameters:
function (callable) – callable mapping a 1D array of inputs to a 1D array of outputs. With
method='jax'it must be written withjax.numpy.fiducial (array-like) – point where the Jacobian is evaluated.
method (
string) –'fd'for central finite differences or'jax'for forward-mode automatic differentiation (requires jax).step (
float) – relative finite-difference step; the absolute step for input \(x_\alpha\) isstep * max(|x_alpha|, 1). Ignored withmethod='jax'.
- Returns:
the Jacobian matrix, shape
(n_outputs, n_inputs).- Return type: