# Licensed under a 3-clause BSD style license - see LICENSE.rst
from __future__ import absolute_import, division, print_function, unicode_literals
import numpy as np
from .likelihood import Likelihood
__all__ = ["optimize_sherpa", "covariance_sherpa"]
def get_sherpa_optimizer(name):
from sherpa.optmethods import LevMar, NelderMead, MonCar, GridSearch
return {
"levmar": LevMar,
"simplex": NelderMead,
"moncar": MonCar,
"gridsearch": GridSearch,
}[name]()
class SherpaLikelihood(Likelihood):
"""Likelihood function interface for Sherpa."""
def fcn(self, factors):
self.parameters.set_parameter_factors(factors)
return self.function(self.parameters), 0
[docs]def optimize_sherpa(parameters, function, **kwargs):
"""Sherpa optimization wrapper method.
Parameters
----------
parameters : `~gammapy.utils.modeling.Parameters`
Parameter list with starting values.
function : callable
Likelihood function
**kwargs : dict
Options passed to the optimizer instance.
Returns
-------
result : (factors, info, optimizer)
Tuple containing the best fit factors, some info and the optimizer instance.
"""
method = kwargs.pop("method", "simplex")
optimizer = get_sherpa_optimizer(method)
optimizer.config.update(kwargs)
pars = [par.factor for par in parameters.parameters]
parmins = [par.factor_min for par in parameters.parameters]
parmaxes = [par.factor_max for par in parameters.parameters]
statfunc = SherpaLikelihood(function, parameters)
with np.errstate(invalid="ignore"):
result = optimizer.fit(
statfunc=statfunc.fcn, pars=pars, parmins=parmins, parmaxes=parmaxes
)
factors = result[1]
info = {"success": result[0], "message": result[3], "nfev": result[4]["nfev"]}
return factors, info, optimizer
[docs]def covariance_sherpa():
raise NotImplementedError