Modeller

class lsst.multiprofit.Modeller(logger: Logger | None = None)

Bases: object

Fit lsst.gauss2d.fit Model instances using Python optimizers.

Parameters:

loggerlogging.Logger

The logger. Defaults to calling _getlogger.

Methods Summary

compute_variances(model[, use_diag_only, ...])	Compute model free parameter variances from the inverse Hessian.
fit_gaussians_linear(gaussians_linear, ...)	Fit normalizations for a Gaussian mixture model.
fit_model(model[, fitinputs, printout, config])	Fit a model with a nonlinear optimizer.
fit_model_linear(model[, idx_obs, ...])	Fit a model's linear parameters (integrals).
make_components_linear(component_mixture)	Make a list of fixed Gaussian components from a ComponentMixture.

Methods Documentation

static compute_variances(model: ModelD | ModelF, use_diag_only: bool = False, use_svd: bool = False, **kwargs: Any) → ndarray

Compute model free parameter variances from the inverse Hessian.

Parameters:

model

The model to compute parameter variances for.

use_diag_only

Whether to use diagonal terms only, i.e. ignore covariance.

use_svd

Whether to use singular value decomposition to compute the inverse Hessian.

**kwargs

Additional keyword arguments to pass to model.compute_hessian.

Returns:

variances

The free parameter variances.

static fit_gaussians_linear(gaussians_linear: LinearGaussians, observation: ObservationD, psf_model: PsfModel = None, fit_methods: dict[str, dict[str, Any]] | None = None, plot: bool = False) → dict[str, lsst.multiprofit.modeller.FitResult]

Fit normalizations for a Gaussian mixture model.

Parameters:

gaussians_linear

The Gaussian components - fixed or otherwise - to fit.

observation

The observation to fit against.

psf_model

A PSF model for the observation, if fitting sources.

fit_methods

A dictionary of fitting methods to employ, keyed by method name, with a value of a dict of options (kwargs) to pass on. Default is “scipy.optimize.nnls”.

plot

Whether to generate fit residual/diagnostic plots.

Returns:

results

Fit results for each method, keyed by the fit method name.

fit_model(model: ModelD | ModelF, fitinputs: FitInputs | None = None, printout: bool = False, config: ModelFitConfig | None = None, **kwargs: Any) → FitResult

Fit a model with a nonlinear optimizer.

Parameters:

model

The model to fit.

fitinputs

An existing FitInputs with jacobian/residual arrays to reuse.

printout

Whether to print diagnostic information.

config

Configuration settings for model fitting.

**kwargs

Keyword arguments to pass to the optimizer.

Returns:

result

The results from running the fitter.

Notes

The only supported fitter is scipy.optimize.least_squares.

classmethod fit_model_linear(model: ModelD | ModelF, idx_obs: int | Sequence[int] | None = None, ratio_min: float = 0, validate: bool = False, limits_interval_min: float = 0.01, limits_interval_max: float = 1.0) → tuple[numpy.ndarray, numpy.ndarray]

Fit a model’s linear parameters (integrals).

Parameters:

model

The model to fit parameters for.

idx_obs

An index or sequence of indices of observations to fit. The default is to fit all observations.

ratio_min

The minimum ratio of the previous value to set any parameter to. This can prevent setting parameters to zero.

validate

If True, check that the model log-likelihood improves and restore the original parameter values if not.

limits_interval_min

A value 0<=x<limits_interval_max<=1 specifying the lower bound to clip parameter values to, as a ratio of each parameter’s limits.

limits_interval_max

A value 0<=limits_interval_min<x<=1 specifying the upper bound to clip parameter values to, as a ratio of each parameter’s limits.

Returns:

loglike_init

The initial log likelihood if validate is True, otherwise None.

loglike_final

The post-fit log likelihood if validate is True, otherwise None.

Notes

The purpose of limits_interval is to slightly offset parameters from the extrema of their limits. This is typically most useful for integral parameters with a minimum of zero, which might otherwise be stuck at zero in a subsequent nonlinear fit.

static make_components_linear(component_mixture: ComponentMixture) → list[lsst.gauss2d.fit._gauss2d_fit.GaussianComponent]

Make a list of fixed Gaussian components from a ComponentMixture.

Parameters:

component_mixture

A component mixture to create a component list for.

Returns:

gaussians

A list of Gaussians components with fixed parameters and values matching those in the original component mixture.