parameter_errors

nyx.infer.uncertainty.parameter_errors(fitted, residuals_fn, *, batch_size=8, reduced_chi2=False, rcond=None)[source]

1-sigma errors on every trainable Parameter, from the Gauss-Newton covariance.

Parameters:

  • fitted (pytree) – Model to linearise about.

  • residuals_fn (callable) – (model) -> residuals.

  • batch_size (int, optional) – Jacobian columns per kernel launch.

  • reduced_chi2 (bool, optional) – Whether to apply the (r^T r) / (m - n) factor. False assumes residuals pre-scaled by 1-sigma uncertainties, so cov = pinv(J^T J).

  • rcond (float or None, optional) – Pseudoinverse cutoff, relative to the largest singular value of J. Defaults to n * eps of the Jacobian’s dtype; a direction below that is treated as unidentifiable and given zero error.

Returns:

sqrt(diag(pinv(J^T J))), shaped like the trainable subset of fitted.

Return type:

pytree