Source code for gammapy.catalog.gammacat

# Licensed under a 3-clause BSD style license - see LICENSE.rst
"""Gammacat open TeV source catalog.

https://github.com/gammapy/gamma-cat
"""
import logging
import numpy as np
from astropy import units as u
from astropy.table import Table
from gammapy.modeling.models import Model, SkyModel
from gammapy.spectrum import FluxPoints
from gammapy.utils.scripts import make_path
from .core import SourceCatalog, SourceCatalogObject

__all__ = ["SourceCatalogGammaCat", "SourceCatalogObjectGammaCat"]

log = logging.getLogger(__name__)


[docs]class SourceCatalogObjectGammaCat(SourceCatalogObject): """One object from the gamma-cat source catalog. Catalog is represented by `~gammapy.catalog.SourceCatalogGammaCat`. """ _source_name_key = "common_name" def __str__(self): return self.info()
[docs] def info(self, info="all"): """Info string. Parameters ---------- info : {'all', 'basic', 'position, 'model'} Comma separated list of options """ if info == "all": info = "basic,position,model" ss = "" ops = info.split(",") if "basic" in ops: ss += self._info_basic() if "position" in ops: ss += self._info_position() if "model" in ops: ss += self._info_morph() ss += self._info_spectral_fit() ss += self._info_spectral_points() return ss
def _info_basic(self): """Print basic info.""" d = self.data return ( f"\n*** Basic info ***\n\n" f"Catalog row index (zero-based): {self.row_index}\n" f"Common name: {self.name}\n" f"Gamma names: {d.gamma_names}\n" f"Fermi names: {d.fermi_names}\n" f"Other names: {d.other_names}\n" f"Location: {d.where}\n" f"Class: {d.classes}\n\n" f"TeVCat ID: {d.tevcat_id}\n" f"TeVCat 2 ID: {d.tevcat2_id}\n" f"TeVCat name: {d.tevcat_name}\n\n" f"TGeVCat ID: {d.tgevcat_id}\n" f"TGeVCat name: {d.tgevcat_name}\n\n" f"Discoverer: {d.discoverer}\n" f"Discovery date: {d.discovery_date}\n" f"Seen by: {d.seen_by}\n" f"Reference: {d.reference_id}\n" ) def _info_position(self): """Print position info.""" d = self.data return ( f"\n*** Position info ***\n\n" f"SIMBAD:\n" f"RA: {d.ra:.3f}\n" f"DEC: {d.dec:.3f}\n" f"GLON: {d.glon:.3f}\n" f"GLAT: {d.glat:.3f}\n" f"\nMeasurement:\n" f"RA: {d.pos_ra:.3f}\n" f"DEC: {d.pos_dec:.3f}\n" f"GLON: {d.pos_glon:.3f}\n" f"GLAT: {d.pos_glat:.3f}\n" f"Position error: {d.pos_err:.3f}\n" ) def _info_morph(self): """Print morphology info.""" d = self.data return ( f"\n*** Morphology info ***\n\n" f"Morphology model type: {d.morph_type}\n" f"Sigma: {d.morph_sigma:.3f}\n" f"Sigma error: {d.morph_sigma_err:.3f}\n" f"Sigma2: {d.morph_sigma2:.3f}\n" f"Sigma2 error: {d.morph_sigma2_err:.3f}\n" f"Position angle: {d.morph_pa:.3f}\n" f"Position angle error: {d.morph_pa_err:.3f}\n" f"Position angle frame: {d.morph_pa_frame}\n" ) def _info_spectral_fit(self): """Print spectral info.""" d = self.data ss = f"\n*** Spectral info ***\n\n" ss += f"Significance: {d.significance:.3f}\n" ss += f"Livetime: {d.livetime:.3f}\n" spec_type = d["spec_type"] ss += f"\nSpectrum type: {spec_type}\n" if spec_type == "pl": ss += f"norm: {d.spec_pl_norm:.3} +- {d.spec_pl_norm_err:.3} (stat) +- {d.spec_pl_norm_err_sys:.3} (sys) cm-2 s-1 TeV-1\n" ss += f"index: {d.spec_pl_index:.3} +- {d.spec_pl_index_err:.3} (stat) +- {d.spec_pl_index_err_sys:.3} (sys)\n" ss += f"reference: {d.spec_pl_e_ref:.3}\n" elif spec_type == "pl2": ss += f"flux: {d.spec_pl2_flux.value:.3} +- {d.spec_pl2_flux_err.value:.3} (stat) +- {d.spec_pl2_flux_err_sys.value:.3} (sys) cm-2 s-1\n" ss += f"index: {d.spec_pl2_index:.3} +- {d.spec_pl2_index_err:.3} (stat) +- {d.spec_pl2_index_err_sys:.3} (sys)\n" ss += f"e_min: {d.spec_pl2_e_min:.3}\n" ss += f"e_max: {d.spec_pl2_e_max:.3}\n" elif spec_type == "ecpl": ss += f"norm: {d.spec_ecpl_norm.value:.3g} +- {d.spec_ecpl_norm_err.value:.3g} (stat) +- {d.spec_ecpl_norm_err_sys.value:.03g} (sys) cm-2 s-1 TeV-1\n" ss += f"index: {d.spec_ecpl_index:.3} +- {d.spec_ecpl_index_err:.3} (stat) +- {d.spec_ecpl_index_err_sys:.3} (sys)\n" ss += f"e_cut: {d.spec_ecpl_e_cut.value:.3} +- {d.spec_ecpl_e_cut_err.value:.3} (stat) +- {d.spec_ecpl_e_cut_err_sys.value:.3} (sys) TeV\n" ss += f"reference: {d.spec_ecpl_e_ref:.3}\n" elif spec_type == "none": pass else: raise ValueError(f"Invalid spec_type: {spec_type}") ss += f"\nEnergy range: ({d.spec_erange_min.value:.3}, {d.spec_erange_max.value:.3}) TeV\n" ss += f"theta: {d.spec_theta:.3}\n" ss += "\n\nDerived fluxes:\n" ss += f"Spectral model norm (1 TeV): {d.spec_dnde_1TeV:.3} +- {d.spec_dnde_1TeV_err:.3} (stat) cm-2 s-1 TeV-1\n" ss += f"Integrated flux (>1 TeV): {d.spec_flux_1TeV.value:.3} +- {d.spec_flux_1TeV_err.value:.3} (stat) cm-2 s-1\n" ss += f"Integrated flux (>1 TeV): {d.spec_flux_1TeV_crab:.3f} +- {d.spec_flux_1TeV_crab_err:.3f} (% Crab)\n" ss += f"Integrated flux (1-10 TeV): {d.spec_eflux_1TeV_10TeV.value:.3} +- {d.spec_eflux_1TeV_10TeV_err.value:.3} (stat) erg cm-2 s-1\n" return ss def _info_spectral_points(self): """Print spectral points info.""" d = self.data ss = "\n*** Spectral points ***\n\n" ss += f"SED reference ID: {d.sed_reference_id}\n" ss += f"Number of spectral points: {d.sed_n_points}\n" ss += f"Number of upper limits: {d.sed_n_ul}\n\n" flux_points = self.flux_points if flux_points is None: ss += "\nNo spectral points available for this source." else: lines = flux_points.table_formatted.pformat(max_width=-1, max_lines=-1) ss += "\n".join(lines) return ss + "\n"
[docs] def spectral_model(self): """Source spectral model (`~gammapy.modeling.models.SpectralModel`). Parameter errors are statistical errors only. """ data = self.data spec_type = data["spec_type"] if spec_type == "pl": tag = "PowerLawSpectralModel" pars = { "amplitude": data["spec_pl_norm"], "index": data["spec_pl_index"], "reference": data["spec_pl_e_ref"], } errs = { "amplitude": data["spec_pl_norm_err"], "index": data["spec_pl_index_err"], } elif spec_type == "pl2": e_max = data["spec_pl2_e_max"] DEFAULT_E_MAX = u.Quantity(1e5, "TeV") if np.isnan(e_max.value): e_max = DEFAULT_E_MAX tag = "PowerLaw2SpectralModel" pars = { "amplitude": data["spec_pl2_flux"], "index": data["spec_pl2_index"], "emin": data["spec_pl2_e_min"], "emax": e_max, } errs = { "amplitude": data["spec_pl2_flux_err"], "index": data["spec_pl2_index_err"], } elif spec_type == "ecpl": tag = "ExpCutoffPowerLawSpectralModel" pars = { "amplitude": data["spec_ecpl_norm"], "index": data["spec_ecpl_index"], "lambda_": 1.0 / data["spec_ecpl_e_cut"], "reference": data["spec_ecpl_e_ref"], } errs = { "amplitude": data["spec_ecpl_norm_err"], "index": data["spec_ecpl_index_err"], "lambda_": data["spec_ecpl_e_cut_err"] / data["spec_ecpl_e_cut"] ** 2, } elif spec_type == "none": return None else: raise ValueError(f"Invalid spec_type: {spec_type}") model = Model.create(tag, **pars) model.parameters.set_error(**errs) return model
[docs] def spatial_model(self): """Source spatial model (`~gammapy.modeling.models.SpatialModel`). TODO: add parameter errors! """ d = self.data morph_type = d["morph_type"] pars = {"lon_0": d["glon"], "lat_0": d["glat"], "frame": "galactic"} errs = { "lat_0": self.data["pos_err"], "lon_0": self.data["pos_err"] / np.cos(self.data["glat"]), } if morph_type == "point": tag = "PointSpatialModel" elif morph_type == "gauss": # TODO: add infos back once model support elongation # pars['x_stddev'] = d['morph_sigma'] # pars['y_stddev'] = d['morph_sigma'] # if not np.isnan(d['morph_sigma2']): # pars['y_stddev'] = d['morph_sigma2'] # if not np.isnan(d['morph_pa']): # # TODO: handle reference frame for rotation angle # pars['theta'] = Angle(d['morph_pa'], 'deg').rad tag = "GaussianSpatialModel" pars["sigma"] = d["morph_sigma"] elif morph_type == "shell": tag = "ShellSpatialModel" # TODO: probably we shouldn't guess a shell width here! pars["radius"] = 0.8 * d["morph_sigma"] pars["width"] = 0.2 * d["morph_sigma"] elif morph_type == "none": return None else: raise ValueError(f"Invalid morph_type: {morph_type!r}") model = Model.create(tag, **pars) model.parameters.set_error(**errs) return model
[docs] def sky_model(self): """Source sky model (`~gammapy.modeling.models.SkyModel`).""" return SkyModel( spatial_model=self.spatial_model(), spectral_model=self.spectral_model(), name=self.name, )
def _add_source_meta(self, table): """Copy over some info to table.meta""" d = self.data m = table.meta m["origin"] = "Data from gamma-cat" m["source_id"] = d["source_id"] m["common_name"] = d["common_name"] m["reference_id"] = d["reference_id"] @property def flux_points(self): """Differential flux points (`~gammapy.spectrum.FluxPoints`).""" d = self.data table = Table() table.meta["SED_TYPE"] = "dnde" self._add_source_meta(table) valid = np.isfinite(d["sed_e_ref"].value) if valid.sum() == 0: return None table["e_ref"] = d["sed_e_ref"] table["e_min"] = d["sed_e_min"] table["e_max"] = d["sed_e_max"] table["dnde"] = d["sed_dnde"] table["dnde_err"] = d["sed_dnde_err"] table["dnde_errn"] = d["sed_dnde_errn"] table["dnde_errp"] = d["sed_dnde_errp"] table["dnde_ul"] = d["sed_dnde_ul"] # Only keep rows that actually contain information table = table[valid] # Only keep columns that actually contain information def _del_nan_col(table, colname): if np.isfinite(table[colname]).sum() == 0: del table[colname] for colname in table.colnames: _del_nan_col(table, colname) return FluxPoints(table)
[docs]class SourceCatalogGammaCat(SourceCatalog): """Gammacat open TeV source catalog. See: https://github.com/gammapy/gamma-cat One source is represented by `~gammapy.catalog.SourceCatalogObjectGammaCat`. Parameters ---------- filename : str Path to the gamma-cat fits file. Examples -------- Load the catalog data: >>> import astropy.units as u >>> from gammapy.catalog import SourceCatalogGammaCat >>> cat = SourceCatalogGammaCat() Access a source by name: >>> source = cat['Vela Junior'] Access source spectral data and plot it: >>> energy_range = [1, 10] * u.TeV >>> source.spectral_model().plot(energy_range) >>> source.spectral_model().plot_error(energy_range) >>> source.flux_points.plot() """ name = "gamma-cat" description = "An open catalog of gamma-ray sources" source_object_class = SourceCatalogObjectGammaCat def __init__(self, filename="$GAMMAPY_DATA/catalogs/gammacat/gammacat.fits.gz"): filename = make_path(filename) table = Table.read(filename, hdu=1) source_name_key = "common_name" source_name_alias = ("other_names", "gamma_names") super().__init__( table=table, source_name_key=source_name_key, source_name_alias=source_name_alias, )