Source code for gammapy.spectrum.results

# 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 astropy.table import Table, Column
import astropy.units as u
from .butterfly import SpectrumButterfly
from ..spectrum import CountsSpectrum, models
from ..extern.bunch import Bunch
from ..utils.scripts import read_yaml, make_path
from ..utils.energy import EnergyBounds

__all__ = [
    'SpectrumFitResult',
    'SpectrumResult',
]


[docs]class SpectrumFitResult(object): """Class representing the result of a spectral fit TODO: This only supports WStat fits at the moment. Find a solution to display also Cash fits (flag, subclass). Parameters ---------- model : `~gammapy.spectrum.models.SpectralModel` Best-fit model fit_range : `~astropy.units.Quantity` Energy range of the spectral fit statname : str, optional Statistic used for the fit statval : float, optional Final fit statistic stat_per_bin : float, optional Fit statistic value per bin npred_src : array-like, optional Source counts predicted by the fit npred_bkg : array-like, optional Background counts predicted by the fit background_model : `~gammapy.spectrum.SpectralModel` Best-fit background model flux_at_1TeV : dict, optional Flux for the fitted model at 1 TeV flux_at_1TeV_err : dict, optional Error on the flux for the fitted model at 1 TeV obs : `~gammapy.spectrum.SpectrumObservation` Input data used for the fit """ def __init__(self, model, fit_range=None, statname=None, statval=None, stat_per_bin=None, npred_src=None, npred_bkg=None, background_model=None, fluxes=None, flux_errors=None, obs=None): self.model = model self.fit_range = fit_range self.statname = statname self.statval = statval self.stat_per_bin = stat_per_bin self.npred_src = npred_src self.npred_bkg = npred_bkg self.background_model = background_model self.fluxes = fluxes self.flux_errors = flux_errors self.obs = obs @classmethod
[docs] def from_yaml(cls, filename): """Create cls from YAML file Parameters ---------- filename : str, Path File to read """ filename = make_path(filename) val = read_yaml(str(filename)) return cls.from_dict(val)
[docs] def to_yaml(self, filename, mode='w'): """Write YAML file Parameters ---------- filename : str File to write mode : str Write mode """ import yaml d = self.to_dict() val = yaml.safe_dump(d, default_flow_style=False) with open(str(filename), mode) as outfile: outfile.write(val)
[docs] def to_dict(self): """Convert to dict""" val = dict() val['model'] = self.model.to_dict() if self.fit_range is not None: val['fit_range'] = dict(min=self.fit_range[0].value, max=self.fit_range[1].value, unit=str(self.fit_range.unit)) if self.statval is not None: val['statval'] = float(self.statval) if self.statname is not None: val['statname'] = self.statname return val
@classmethod
[docs] def from_dict(cls, val): modeldict = val['model'] model = models.SpectralModel.from_dict(modeldict) try: erange = val['fit_range'] energy_range = (erange['min'], erange['max']) * u.Unit(erange['unit']) except KeyError: energy_range = None try: fl = val['fluxes'] except KeyError: fluxes = None flux_errors = None else: fluxes = Bunch() flux_errors = Bunch() for flu in fl: fluxes[flu] = fl[flu]['value'] * u.Unit(fl[flu]['unit']) flux_errors[flu] = fl[flu]['error'] * u.Unit(fl[flu]['unit']) return cls(model=model, fit_range=energy_range, fluxes=fluxes, flux_errors=flux_errors)
# TODO: rather add this to ParameterList?
[docs] def to_table(self, energy_unit='TeV', flux_unit='cm-2 s-1 TeV-1', **kwargs): """Convert to `~astropy.table.Table` Produce overview table containing the most important parameters """ t = Table() t['model'] = [self.model.__class__.__name__] for par_name, value in self.model.parameters._ufloats.items(): val = value.n err = value.s # Apply correction factor for units # TODO: Refactor current_unit = self.model.parameters[par_name].unit if current_unit.is_equivalent(energy_unit): factor = current_unit.to(energy_unit) col_unit = energy_unit elif current_unit.is_equivalent(1 / u.Unit(energy_unit)): factor = current_unit.to(1 / u.Unit(energy_unit)) col_unit = 1 / u.Unit(energy_unit) elif current_unit.is_equivalent(flux_unit): factor = current_unit.to(flux_unit) col_unit = flux_unit elif current_unit.is_equivalent(u.dimensionless_unscaled): factor = 1 col_unit = current_unit else: raise ValueError(current_unit) t[par_name] = Column( data=np.atleast_1d(val * factor), unit=col_unit, **kwargs) t['{}_err'.format(par_name)] = Column( data=np.atleast_1d(err * factor), unit=col_unit, **kwargs) t['fit_range'] = Column( data=[self.fit_range.to(energy_unit)], unit=energy_unit, **kwargs) return t
def __str__(self): """ Summary info string. """ info = '\nFit result info \n' info += '--------------- \n' info += 'Model: {} \n'.format(self.model) if self.statval is not None: info += '\nStatistic: {0:.3f} ({1})'.format(self.statval, self.statname) if self.fit_range is not None: info += '\nFit Range: {}'.format(self.fit_range) info += '\n' return info
[docs] def info(self): """ Print summary info. """ print(str(self))
[docs] def butterfly(self, energy=None, flux_unit='TeV-1 cm-2 s-1'): """ Compute butterfly. Parameters ---------- energy : `~astropy.units.Quantity`, optional Energies at which to evaluate the butterfly. flux_unit : str Flux unit for the butterfly. Returns ------- butterfly : `~gammapy.spectrum.SpectrumButterfly` Butterfly object. """ if energy is None: energy = EnergyBounds.equal_log_spacing(self.fit_range[0], self.fit_range[1], 100) flux, flux_err = self.model.evaluate_error(energy) butterfly = SpectrumButterfly() butterfly['energy'] = energy butterfly['flux'] = flux.to(flux_unit) butterfly['flux_lo'] = flux - flux_err.to(flux_unit) butterfly['flux_hi'] = flux + flux_err.to(flux_unit) return butterfly
@property def expected_source_counts(self): """`~gammapy.spectrum.CountsSpectrum` of predicted source counts """ energy = self.obs.on_vector.energy data = self.npred_src * u.ct return CountsSpectrum(data=data, energy_lo=energy.lo, energy_hi=energy.hi) @property def expected_background_counts(self): """`~gammapy.spectrum.CountsSpectrum` of predicted background counts """ try: energy = self.obs.e_reco data = self.npred_bkg * u.ct return CountsSpectrum(data=data, energy_hi=energy.upper_bounds, energy_lo=energy.lower_bounds) except TypeError: return None @property def expected_on_counts(self): """`~gammapy.spectrum.CountsSpectrum` of predicted on counts """ mu_on = self.expected_source_counts.copy() mu_on.data.data += self.expected_background_counts.data.data return mu_on @property def residuals(self): """Residuals Prediced on counts - expected on counts """ resspec = self.expected_on_counts.copy() resspec.data.data -= self.obs.on_vector.data.data return resspec
[docs] def plot(self, **kwargs): """Standard debug plot. Plot ON counts in comparison to model. """ ax0, ax1 = get_plot_axis(**kwargs) self.plot_counts(ax0) self.plot_residuals(ax1) return ax0, ax1
[docs] def plot_counts(self, ax): """Plot predicted and detected counts.""" self.expected_source_counts.plot(ax=ax, label='mu_source') self.expected_background_counts.plot(ax=ax, label='mu_background', energy_unit='TeV') self.expected_on_counts.plot(ax=ax, label='mu_on', energy_unit='TeV') self.obs.on_vector.plot(ax=ax, label='n_on', show_poisson_errors=True, fmt='.', energy_unit='TeV') ax.legend(numpoints=1) ax.set_title('')
[docs] def plot_residuals(self, ax): """Plot residuals.""" self.residuals.plot(ax=ax, ecolor='black', fmt='none') xx = ax.get_xlim() yy = [0, 0] ax.plot(xx, yy, color='black') ymax = 1.4 * max(self.residuals.data.data.value) ax.set_ylim(-ymax, ymax) xmin = self.fit_range.to('TeV').value[0] * 0.8 xmax = self.fit_range.to('TeV').value[1] * 1.2 ax.set_xlim(xmin, xmax) ax.set_xlabel('Energy [{}]'.format('TeV')) ax.set_ylabel('ON (Predicted - Detected)')
[docs]class SpectrumResult(object): """Class holding all results of a spectral analysis Best fit model, flux points Parameters ---------- model : `~gammapy.spectrum.models.SpectralModel` Best Fit model points : `~gammapy.spectrum.FluxPoints`, optional Flux points """ def __init__(self, model, points): self.model = model self.points = points @property def flux_point_residuals(self): """Residuals Defined as ``(points - model) / model`` Returns ------- residuals : np.array Residuals residuals_err : np.array Residuals error """ e_ref = self.points.table['e_ref'].quantity points = self.points.table['dnde'].quantity points_err = self.points.get_flux_err() # Deal with asymetric errors if type(points_err) == tuple: points_err = np.sqrt(points_err[0] * points_err[1]) model_val = self.model(e_ref) residuals = ((points - model_val) / model_val).to('') residuals_err = (points_err / model_val).to('') return residuals, residuals_err
[docs] def plot(self, energy_range, energy_unit='TeV', flux_unit='cm-2 s-1 TeV-1', energy_power=0, fit_kwargs=dict(), butterfly_kwargs=dict(), point_kwargs=dict(), fig_kwargs=dict()): """Plot spectrum Plot best fit model, flux points and residuals Parameters ---------- energy_range : `~astropy.units.Quantity` Energy range for the plot energy_unit : str, `~astropy.units.Unit`, optional Unit of the energy axis flux_unit : str, `~astropy.units.Unit`, optional Unit of the flux axis energy_power : int Power of energy to multiply flux axis with fit_kwargs : dict, optional forwarded to :func:`gammapy.spectrum.models.SpectralModel.plot` butterfly_kwargs : dict, optional forwarded to :func:`gammapy.spectrum.SpectralModel.plot_error` point_kwargs : dict, optional forwarded to :func:`gammapy.spectrum.FluxPoints.plot` fig_kwargs : dict, optional forwarded to :func:`matplotlib.pyplot.figure` Returns ------- ax0 : `~matplotlib.axes.Axes` Spectrum plot axis ax1 : `~matplotlib.axes.Axes` Residuals plot axis """ import matplotlib.pyplot as plt ax0, ax1 = get_plot_axis(**fig_kwargs) ax0.set_yscale('log') common_kwargs = dict( energy_unit=energy_unit, flux_unit=flux_unit, energy_power=energy_power) fit_kwargs.update(common_kwargs) point_kwargs.update(common_kwargs) butterfly_kwargs.update(common_kwargs) self.model.plot(energy_range=energy_range, ax=ax0, **fit_kwargs) self.model.plot_error(energy_range=energy_range, ax=ax0, **butterfly_kwargs) self.points.plot(ax=ax0, **point_kwargs) point_kwargs.pop('flux_unit') point_kwargs.pop('energy_power') ax0.set_xlabel('') self._plot_residuals(ax=ax1, **point_kwargs) return ax0, ax1
def _plot_residuals(self, ax=None, energy_unit='TeV', **kwargs): """Plot residuals Parameters ---------- ax : `~matplotlib.axes.Axes`, optional Axis energy_unit : str, `~astropy.units.Unit`, optional Unit of the energy axis Returns ------- ax : `~matplotlib.axes.Axes`, optional Axis """ import matplotlib.pyplot as plt ax = plt.gca() if ax is None else ax kwargs.setdefault('fmt', '.') y, y_err = self.flux_point_residuals x = self.points.e_ref x = x.to(energy_unit).value ax.errorbar(x, y, yerr=y_err, **kwargs) ax.axhline(0, color='black') ax.set_xlabel('Energy [{}]'.format(energy_unit)) ax.set_ylabel('Residuals') return ax
def get_plot_axis(**kwargs): """Axis setup used for standard plots kwargs are forwarded to plt.figure() Returns ------- ax0 : `~matplotlib.axes.Axes` Main plot ax1 : `~matplotlib.axes.Axes` Residuals """ from matplotlib import gridspec import matplotlib.pyplot as plt fig = plt.figure(**kwargs) gs = gridspec.GridSpec(5, 1) ax0 = plt.subplot(gs[:-2, :]) ax1 = plt.subplot(gs[3, :], sharex=ax0) gs.update(hspace=0.1) plt.setp(ax0.get_xticklabels(), visible=False) ax0.set_xscale('log') ax1.set_xscale('log') return ax0, ax1