Source code for gammapy.makers.background.fov
# Licensed under a 3-clause BSD style license - see LICENSE.rst
"""FoV background estimation."""
import logging
import numpy as np
from gammapy.maps import Map, RegionGeom
from gammapy.modeling import Fit
from gammapy.modeling.models import FoVBackgroundModel, Model
from ..core import Maker
__all__ = ["FoVBackgroundMaker"]
log = logging.getLogger(__name__)
[docs]
class FoVBackgroundMaker(Maker):
"""Normalize template background on the whole field-of-view.
The dataset background model can be simply scaled (method="scale") or fitted
(method="fit") on the dataset counts.
The normalization is performed outside the exclusion mask that is passed on
init. This also internally takes into account the dataset fit mask.
If a SkyModel is set on the input dataset its parameters
are frozen during the FoV re-normalization.
If the requirement (greater than) of either min_counts or min_npred_background
is not satisfied, the background will not be normalised.
Parameters
----------
method : {'scale', 'fit'}
The normalization method to be applied. Default 'scale'.
exclusion_mask : `~gammapy.maps.WcsNDMap`
Exclusion mask.
spectral_model : SpectralModel or str
Reference norm spectral model to use for the `FoVBackgroundModel`, if
none is defined on the dataset. By default, use pl-norm.
min_counts : int
Minimum number of counts, or residuals counts if a SkyModel is set,
required outside the exclusion region.
min_npred_background : float
Minimum number of predicted background counts required outside the
exclusion region.
"""
tag = "FoVBackgroundMaker"
available_methods = ["fit", "scale"]
def __init__(
self,
method="scale",
exclusion_mask=None,
spectral_model="pl-norm",
min_counts=0,
min_npred_background=0,
fit=None,
):
self.method = method
self.exclusion_mask = exclusion_mask
self.min_counts = min_counts
self.min_npred_background = min_npred_background
if isinstance(spectral_model, str):
spectral_model = Model.create(tag=spectral_model, model_type="spectral")
if not spectral_model.is_norm_spectral_model:
raise ValueError("Spectral model must be a norm spectral model")
self.default_spectral_model = spectral_model
if fit is None:
fit = Fit()
self.fit = fit
@property
def method(self):
"""Method property."""
return self._method
@method.setter
def method(self, value):
"""Method setter."""
if value not in self.available_methods:
raise ValueError(
f"Not a valid method for FoVBackgroundMaker: {value}."
f" Choose from {self.available_methods}"
)
self._method = value
[docs]
def make_default_fov_background_model(self, dataset):
"""Add FoV background model to the model definition.
Parameters
----------
dataset : `~gammapy.datasets.MapDataset`
Input map dataset.
Returns
-------
dataset : `~gammapy.datasets.MapDataset`
Map dataset including background model.
"""
bkg_model = FoVBackgroundModel(
dataset_name=dataset.name, spectral_model=self.default_spectral_model.copy()
)
if dataset.models is None:
dataset.models = bkg_model
else:
dataset.models = dataset.models + bkg_model
return dataset
[docs]
def make_exclusion_mask(self, dataset):
"""Project input exclusion mask to dataset geometry.
Parameters
----------
dataset : `~gammapy.datasets.MapDataset`
Input map dataset.
Returns
-------
mask : `~gammapy.maps.WcsNDMap`
Projected exclusion mask.
"""
geom = dataset._geom
if self.exclusion_mask:
mask = self.exclusion_mask.interp_to_geom(geom=geom)
else:
mask = Map.from_geom(geom=geom, data=1, dtype=bool)
return mask
def _make_masked_summed_counts(self, dataset):
"""Compute the sums of the counts, npred, and background maps within the mask."""
npred = dataset.npred()
mask = dataset.mask & ~np.isnan(npred)
count_tot = dataset.counts.data[mask].sum()
npred_tot = npred.data[mask].sum()
bkg_tot = dataset.npred_background().data[mask].sum()
return {
"counts": count_tot,
"npred": npred_tot,
"bkg": bkg_tot,
}
def _verify_requirements(self, dataset):
"""Verify that the requirements of min_counts and min_npred_background are satisfied."""
total = self._make_masked_summed_counts(dataset)
not_bkg_tot = total["npred"] - total["bkg"]
value = (total["counts"] - not_bkg_tot) / total["bkg"]
if not np.isfinite(value):
log.warning(
f"FoVBackgroundMaker failed. Non-finite normalisation value for {dataset.name}. "
"Setting mask to False."
)
return False
elif total["bkg"] <= self.min_npred_background:
log.warning(
f"FoVBackgroundMaker failed. Only {int(total['bkg'])} background"
f" counts outside exclusion mask for {dataset.name}. "
"Setting mask to False."
)
return False
elif total["counts"] <= self.min_counts:
log.warning(
f"FoVBackgroundMaker failed. Only {int(total['counts'])} counts "
f"outside exclusion mask for {dataset.name}. "
"Setting mask to False."
)
return False
elif total["counts"] - not_bkg_tot <= 0:
log.warning(
f"FoVBackgroundMaker failed. Negative residuals counts for {dataset.name}. "
"Setting mask to False."
)
return False
else:
return True
[docs]
def run(self, dataset, observation=None):
"""Run FoV background maker.
Parameters
----------
dataset : `~gammapy.datasets.MapDataset`
Input map dataset.
"""
if isinstance(dataset.counts.geom, RegionGeom):
raise TypeError(
"FoVBackgroundMaker does not support region based datasets."
)
mask_fit = dataset.mask_fit
if mask_fit:
dataset.mask_fit *= self.make_exclusion_mask(dataset)
else:
dataset.mask_fit = self.make_exclusion_mask(dataset)
if dataset.background_model is None:
dataset = self.make_default_fov_background_model(dataset)
if self._verify_requirements(dataset) is True:
if self.method == "fit":
dataset = self.make_background_fit(dataset)
else:
# always scale the background first
dataset = self.make_background_scale(dataset)
else:
dataset.mask_safe.data[...] = False
dataset.mask_fit = mask_fit
return dataset
[docs]
def make_background_fit(self, dataset):
"""Fit the FoV background model on the dataset counts data.
Parameters
----------
dataset : `~gammapy.datasets.MapDataset`
Input dataset.
Returns
-------
dataset : `~gammapy.datasets.MapDataset`
Map dataset with fitted background model.
"""
# freeze all model components not related to background model
models = dataset.models.select(tag="sky-model")
with models.restore_status(restore_values=False):
models.select(tag="sky-model").freeze()
fit_result = self.fit.run(datasets=[dataset])
if not fit_result.success:
log.warning(
f"FoVBackgroundMaker failed. Fit did not converge for {dataset.name}. "
"Setting mask to False."
)
dataset.mask_safe.data[...] = False
return dataset
[docs]
def make_background_scale(self, dataset):
"""Fit the FoV background model on the dataset counts data.
Parameters
----------
dataset : `~gammapy.datasets.MapDataset`
Input dataset.
Returns
-------
dataset : `~gammapy.datasets.MapDataset`
Map dataset with scaled background model.
"""
total = self._make_masked_summed_counts(dataset)
not_bkg_tot = total["npred"] - total["bkg"]
value = (total["counts"] - not_bkg_tot) / total["bkg"]
error = np.sqrt(total["counts"] - not_bkg_tot) / total["bkg"]
dataset.models[f"{dataset.name}-bkg"].spectral_model.norm.value = value
dataset.models[f"{dataset.name}-bkg"].spectral_model.norm.error = error
return dataset
