Source code for gammapy.modeling.fit

# Licensed under a 3-clause BSD style license - see LICENSE.rst
import itertools
import logging
import numpy as np
from astropy.table import Table
from gammapy.utils.deprecation import deprecated
from gammapy.utils.pbar import progress_bar
from .covariance import Covariance
from .iminuit import (
    confidence_iminuit,
    contour_iminuit,
    covariance_iminuit,
    optimize_iminuit,
)
from .scipy import confidence_scipy, optimize_scipy
from .sherpa import optimize_sherpa

__all__ = ["Fit"]

log = logging.getLogger(__name__)


class Registry:
    """Registry of available backends for given tasks.

    Gives users the power to extend from their scripts.
    Used by `Fit` below.

    Not sure if we should call it "backend" or "method" or something else.
    Probably we will code up some methods, e.g. for profile analysis ourselves,
    using scipy or even just Python / Numpy?
    """

    register = {
        "optimize": {
            "minuit": optimize_iminuit,
            "sherpa": optimize_sherpa,
            "scipy": optimize_scipy,
        },
        "covariance": {
            "minuit": covariance_iminuit,
            # "sherpa": covariance_sherpa,
            # "scipy": covariance_scipy,
        },
        "confidence": {
            "minuit": confidence_iminuit,
            # "sherpa": confidence_sherpa,
            "scipy": confidence_scipy,
        },
    }

    @classmethod
    def get(cls, task, backend):
        if task not in cls.register:
            raise ValueError(f"Unknown task {task!r}")

        backend_options = cls.register[task]

        if backend not in backend_options:
            raise ValueError(f"Unknown backend {backend!r} for task {task!r}")

        return backend_options[backend]


registry = Registry()


[docs]class Fit: """Fit class. The fit class provides a uniform interface to multiple fitting backends. Currently available: "minuit", "sherpa" and "scipy" Parameters ---------- backend : {"minuit", "scipy" "sherpa"} Global backend used for fitting, default : minuit optimize_opts : dict Keyword arguments passed to the optimizer. For the `"minuit"` backend see https://iminuit.readthedocs.io/en/stable/reference.html#iminuit.Minuit for a detailed description of the available options. If there is an entry 'migrad_opts', those options will be passed to `iminuit.Minuit.migrad()`. For the `"sherpa"` backend you can from the options: * `"simplex"` * `"levmar"` * `"moncar"` * `"gridsearch"` Those methods are described and compared in detail on http://cxc.cfa.harvard.edu/sherpa/methods/index.html. The available options of the optimization methods are described on the following pages in detail: * http://cxc.cfa.harvard.edu/sherpa/ahelp/neldermead.html * http://cxc.cfa.harvard.edu/sherpa/ahelp/montecarlo.html * http://cxc.cfa.harvard.edu/sherpa/ahelp/gridsearch.html * http://cxc.cfa.harvard.edu/sherpa/ahelp/levmar.html For the `"scipy"` backend the available options are described in detail here: https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.minimize.html covariance_opts : dict Covariance options passed to the given backend. confidence_opts : dict Extra arguments passed to the backend. E.g. `iminuit.Minuit.minos` supports a ``maxcall`` option. For the scipy backend ``confidence_opts`` are forwarded to `~scipy.optimize.brentq`. If the confidence estimation fails, the bracketing interval can be adapted by modifying the the upper bound of the interval (``b``) value. store_trace : bool Whether to store the trace of the fit """ def __init__( self, backend="minuit", optimize_opts=None, covariance_opts=None, confidence_opts=None, store_trace=False, ): self.store_trace = store_trace self.backend = backend if optimize_opts is None: optimize_opts = {"backend": backend} if covariance_opts is None: covariance_opts = {"backend": backend} if confidence_opts is None: confidence_opts = {"backend": backend} self.optimize_opts = optimize_opts self.covariance_opts = covariance_opts self.confidence_opts = confidence_opts self._minuit = None @property @deprecated( "v1.1", message="The IMinuit object is attached to the OptimizeResult object instead.", ) def minuit(self): """Iminuit object""" return self._minuit @staticmethod def _parse_datasets(datasets): from gammapy.datasets import Datasets datasets = Datasets(datasets) return datasets, datasets.parameters
[docs] def run(self, datasets): """Run all fitting steps. Parameters ---------- datasets : `Datasets` or list of `Dataset` Datasets to optimize. Returns ------- fit_result : `FitResult` Fit result """ optimize_result = self.optimize(datasets=datasets) if self.backend not in registry.register["covariance"]: log.warning("No covariance estimate - not supported by this backend.") return FitResult(optimize_result=optimize_result) covariance_result = self.covariance( datasets=datasets, optimize_result=optimize_result ) optimize_result.models.covariance = Covariance( optimize_result.models.parameters, covariance_result.matrix ) return FitResult( optimize_result=optimize_result, covariance_result=covariance_result, )
[docs] def optimize(self, datasets): """Run the optimization. Parameters ---------- datasets : `Datasets` or list of `Dataset` Datasets to optimize. Returns ------- optimize_result : `OptimizeResult` Optimization result """ datasets, parameters = self._parse_datasets(datasets=datasets) datasets.parameters.check_limits() if len(parameters.free_parameters.names) == 0: raise ValueError("No free parameters for fitting") parameters.autoscale() kwargs = self.optimize_opts.copy() backend = kwargs.pop("backend", self.backend) compute = registry.get("optimize", backend) # TODO: change this calling interface! # probably should pass a fit statistic, which has a model, which has parameters # and return something simpler, not a tuple of three things factors, info, optimizer = compute( parameters=parameters, function=datasets.stat_sum, store_trace=self.store_trace, **kwargs, ) if backend == "minuit": self._minuit = optimizer kwargs["method"] = "migrad" trace = Table(info.pop("trace")) if self.store_trace: idx = [ parameters.index(par) for par in parameters.unique_parameters.free_parameters ] unique_names = np.array(datasets.models.parameters_unique_names)[idx] trace.rename_columns(trace.colnames[1:], list(unique_names)) # Copy final results into the parameters object parameters.set_parameter_factors(factors) parameters.check_limits() return OptimizeResult( models=datasets.models.copy(), total_stat=datasets.stat_sum(), backend=backend, method=kwargs.get("method", backend), trace=trace, minuit=optimizer, **info, )
[docs] def covariance(self, datasets, optimize_result=None): """Estimate the covariance matrix. Assumes that the model parameters are already optimised. Parameters ---------- datasets : `Datasets` or list of `Dataset` Datasets to optimize. optimize_result : `OptimizeResult` Optimization result. Can be optionally used to pass the state of the IMinuit object to the covariance estimation. This might save computation time in certain cases. Returns ------- result : `CovarianceResult` Results """ datasets, unique_pars = self._parse_datasets(datasets=datasets) parameters = datasets.models.parameters kwargs = self.covariance_opts.copy() if optimize_result is not None and optimize_result.backend == "minuit": kwargs["minuit"] = optimize_result.minuit backend = kwargs.pop("backend", self.backend) compute = registry.get("covariance", backend) with unique_pars.restore_status(): if self.backend == "minuit": method = "hesse" else: method = "" factor_matrix, info = compute( parameters=unique_pars, function=datasets.stat_sum, **kwargs ) matrix = Covariance.from_factor_matrix( parameters=parameters, matrix=factor_matrix ) datasets.models.covariance = matrix if optimize_result: optimize_result.models.covariance = matrix.data.copy() # TODO: decide what to return, and fill the info correctly! return CovarianceResult( backend=backend, method=method, success=info["success"], message=info["message"], matrix=optimize_result.models.covariance.data, )
[docs] def confidence(self, datasets, parameter, sigma=1, reoptimize=True): """Estimate confidence interval. Extra ``kwargs`` are passed to the backend. E.g. `iminuit.Minuit.minos` supports a ``maxcall`` option. For the scipy backend ``kwargs`` are forwarded to `~scipy.optimize.brentq`. If the confidence estimation fails, the bracketing interval can be adapted by modifying the the upper bound of the interval (``b``) value. Parameters ---------- datasets : `Datasets` or list of `Dataset` Datasets to optimize. parameter : `~gammapy.modeling.Parameter` Parameter of interest sigma : float Number of standard deviations for the confidence level reoptimize : bool Re-optimize other parameters, when computing the confidence region. Returns ------- result : dict Dictionary with keys "errp", 'errn", "success" and "nfev". """ datasets, parameters = self._parse_datasets(datasets=datasets) kwargs = self.confidence_opts.copy() backend = kwargs.pop("backend", self.backend) compute = registry.get("confidence", backend) parameter = parameters[parameter] with parameters.restore_status(): result = compute( parameters=parameters, parameter=parameter, function=datasets.stat_sum, sigma=sigma, reoptimize=reoptimize, **kwargs, ) result["errp"] *= parameter.scale result["errn"] *= parameter.scale return result
[docs] def stat_profile(self, datasets, parameter, reoptimize=False): """Compute fit statistic profile. The method used is to vary one parameter, keeping all others fixed. So this is taking a "slice" or "scan" of the fit statistic. Parameters ---------- datasets : `Datasets` or list of `Dataset` Datasets to optimize. parameter : `~gammapy.modeling.Parameter` Parameter of interest. The specification for the scan, such as bounds and number of values is taken from the parameter object. reoptimize : bool Re-optimize other parameters, when computing the confidence region. Returns ------- results : dict Dictionary with keys "parameter_name_scan", "stat_scan" and "fit_results". The latter contains an empty list, if `reoptimize` is set to False """ datasets, parameters = self._parse_datasets(datasets=datasets) parameter = parameters[parameter] values = parameter.scan_values stats = [] fit_results = [] with parameters.restore_status(): for value in progress_bar(values, desc="Scan values"): parameter.value = value if reoptimize: parameter.frozen = True result = self.optimize(datasets=datasets) stat = result.total_stat fit_results.append(result) else: stat = datasets.stat_sum() stats.append(stat) idx = datasets.parameters.index(parameter) name = datasets.models.parameters_unique_names[idx] return { f"{name}_scan": values, "stat_scan": np.array(stats), "fit_results": fit_results, }
[docs] def stat_surface(self, datasets, x, y, reoptimize=False): """Compute fit statistic surface. The method used is to vary two parameters, keeping all others fixed. So this is taking a "slice" or "scan" of the fit statistic. Caveat: This method can be very computationally intensive and slow See also: `Fit.stat_contour` Parameters ---------- datasets : `Datasets` or list of `Dataset` Datasets to optimize. x, y : `~gammapy.modeling.Parameter` Parameters of interest reoptimize : bool Re-optimize other parameters, when computing the confidence region. Returns ------- results : dict Dictionary with keys "x_values", "y_values", "stat" and "fit_results". The latter contains an empty list, if `reoptimize` is set to False """ datasets, parameters = self._parse_datasets(datasets=datasets) x = parameters[x] y = parameters[y] stats = [] fit_results = [] with parameters.restore_status(): for x_value, y_value in progress_bar( itertools.product(x.scan_values, y.scan_values), desc="Trial values" ): x.value, y.value = x_value, y_value if reoptimize: x.frozen, y.frozen = True, True result = self.optimize(datasets=datasets) stat = result.total_stat fit_results.append(result) else: stat = datasets.stat_sum() stats.append(stat) shape = (len(x.scan_values), len(y.scan_values)) stats = np.array(stats).reshape(shape) if reoptimize: fit_results = np.array(fit_results).reshape(shape) i1, i2 = datasets.parameters.index(x), datasets.parameters.index(y) name_x = datasets.models.parameters_unique_names[i1] name_y = datasets.models.parameters_unique_names[i2] return { f"{name_x}_scan": x.scan_values, f"{name_y}_scan": y.scan_values, "stat_scan": stats, "fit_results": fit_results, }
[docs] def stat_contour(self, datasets, x, y, numpoints=10, sigma=1): """Compute stat contour. Calls ``iminuit.Minuit.mncontour``. This is a contouring algorithm for a 2D function which is not simply the fit statistic function. That 2D function is given at each point ``(par_1, par_2)`` by re-optimising all other free parameters, and taking the fit statistic at that point. Very compute-intensive and slow. Parameters ---------- datasets : `Datasets` or list of `Dataset` Datasets to optimize. x, y : `~gammapy.modeling.Parameter` Parameters of interest numpoints : int Number of contour points sigma : float Number of standard deviations for the confidence level Returns ------- result : dict Dictionary containing the parameter values defining the contour, with the boolean flag "success" and the info objects from ``mncontour``. """ datasets, parameters = self._parse_datasets(datasets=datasets) x = parameters[x] y = parameters[y] i1, i2 = datasets.parameters.index(x), datasets.parameters.index(y) name_x = datasets.models.parameters_unique_names[i1] name_y = datasets.models.parameters_unique_names[i2] with parameters.restore_status(): result = contour_iminuit( parameters=parameters, function=datasets.stat_sum, x=x, y=y, numpoints=numpoints, sigma=sigma, ) x = result["x"] * x.scale y = result["y"] * y.scale return { name_x: x, name_y: y, "success": result["success"], }
class FitStepResult: """Fit result base class""" def __init__(self, backend, method, success, message): self._success = success self._message = message self._backend = backend self._method = method @property def backend(self): """Optimizer backend used for the fit.""" return self._backend @property def method(self): """Optimizer method used for the fit.""" return self._method @property def success(self): """Fit success status flag.""" return self._success @property def message(self): """Optimizer status message.""" return self._message def __repr__(self): return ( f"{self.__class__.__name__}\n\n" f"\tbackend : {self.backend}\n" f"\tmethod : {self.method}\n" f"\tsuccess : {self.success}\n" f"\tmessage : {self.message}\n" ) class CovarianceResult(FitStepResult): """Covariance result object.""" def __init__(self, matrix=None, **kwargs): self._matrix = matrix super().__init__(**kwargs) @property def matrix(self): """Covariance matrix (`~numpy.ndarray`)""" return self._matrix class OptimizeResult(FitStepResult): """Optimize result object.""" def __init__(self, models, nfev, total_stat, trace, minuit=None, **kwargs): self._models = models self._nfev = nfev self._total_stat = total_stat self._trace = trace self._minuit = minuit super().__init__(**kwargs) @property def minuit(self): """Minuit object""" return self._minuit @property def parameters(self): """Best fit parameters""" return self.models.parameters @property def models(self): """Best fit models""" return self._models @property def trace(self): """Parameter trace from the optimisation""" return self._trace @property def nfev(self): """Number of function evaluations.""" return self._nfev @property def total_stat(self): """Value of the fit statistic at minimum.""" return self._total_stat def __repr__(self): str_ = super().__repr__() str_ += f"\tnfev : {self.nfev}\n" str_ += f"\ttotal stat : {self.total_stat:.2f}\n\n" return str_ class FitResult: """Fit result class Parameters ---------- optimize_result : `OptimizeResult` Result of the optimization step. covariance_result : `CovarianceResult` Result of the covariance step. """ def __init__(self, optimize_result=None, covariance_result=None): self._optimize_result = optimize_result self._covariance_result = covariance_result @property def minuit(self): """Minuit object""" return self.optimize_result.minuit # TODO: is the convenience access needed? @property def parameters(self): """Best fit parameters of the optimization step""" return self.optimize_result.parameters # TODO: is the convenience access needed? @property def models(self): """Best fit parameters of the optimization step""" return self.optimize_result.models # TODO: is the convenience access needed? @property def total_stat(self): """Total stat of the optimization step""" return self.optimize_result.total_stat # TODO: is the convenience access needed? @property def trace(self): """Parameter trace of the optimisation step""" return self.optimize_result.trace # TODO: is the convenience access needed? @property def nfev(self): """Number of function evaluations of the optimisation step""" return self.optimize_result.nfev # TODO: is the convenience access needed? @property def backend(self): """Optimizer backend used for the fit.""" return self.optimize_result.backend # TODO: is the convenience access needed? @property def method(self): """Optimizer method used for the fit.""" return self.optimize_result.method # TODO: is the convenience access needed? @property def message(self): """Optimizer status message.""" return self.optimize_result.message @property def success(self): """Total success flag""" success = self.optimize_result.success if self.covariance_result: success &= self.covariance_result.success return success @property def optimize_result(self): """Optimize result""" return self._optimize_result @property def covariance_result(self): """Optimize result""" return self._covariance_result def __repr__(self): str_ = "" if self.optimize_result: str_ += str(self.optimize_result) if self.covariance_result: str_ += str(self.covariance_result) return str_