Trait FitModel

Source

pub trait FitModel {
    // Required method
    fn evaluate(&self, params: &[f64]) -> Result<Vec<f64>, FittingError>;

    // Provided method
    fn analytical_jacobian(
        &self,
        _params: &[f64],
        _free_param_indices: &[usize],
        _y_current: &[f64],
    ) -> Option<FlatMatrix> { ... }
}

Expand description

A model function that can be fitted.

Given parameter values (all params including fixed), computes the model prediction at each data point.

Required Methods§

Source

fn evaluate(&self, params: &[f64]) -> Result<Vec<f64>, FittingError>

Evaluate the model for the given parameters.

On success, returns a vector of model predictions with the same length as the data being fitted. On failure, returns a FittingError indicating that the model could not be evaluated (e.g. a broadening or physics error).

Optimizer semantics: during Levenberg-Marquardt trial steps, an Err is treated as a failed step (increase λ / backtrack). At the initial point or post-convergence, Err is propagated.

Provided Methods§

Source

fn analytical_jacobian( &self, _params: &[f64], _free_param_indices: &[usize], _y_current: &[f64], ) -> Option<FlatMatrix>

Optionally provide an analytical Jacobian.

free_param_indices: indices (into params) of the free parameters, in the same order as the Jacobian columns.

y_current: current model output, i.e. self.evaluate(params). Provided so implementations can compute J analytically from T without an extra evaluate call.

Returns Some(J) where J.get(i, j) = ∂model[i]/∂params[free_param_indices[j]]. The matrix has y_current.len() rows and free_param_indices.len() columns. Return None to fall back to finite-difference Jacobian (the default).

Implementations on Foreign Types§

Source §

impl<M: FitModel + ?Sized> FitModel for &M

Blanket implementation: shared references to any FitModel also implement FitModel, forwarding all calls to the underlying implementation.

This enables NormalizedTransmissionModel<&dyn FitModel> to work when the inner model is borrowed (e.g. in fit_spectrum where the inner model is a local variable wrapped conditionally).

Source §