1
|
|
|
# Licensed under a 3-clause BSD style license - see LICENSE.rst |
2
|
|
|
import numpy as np |
3
|
|
|
from .observation import SpectrumObservation |
4
|
|
|
from .utils import SpectrumEvaluator |
5
|
|
|
from ..utils.fitting import Dataset, Parameters |
6
|
|
|
from ..stats import wstat |
7
|
|
|
|
8
|
|
|
__all__ = ["SpectrumDatasetOnOff"] |
9
|
|
|
|
10
|
|
|
|
11
|
|
|
class SpectrumDatasetOnOff(Dataset): |
12
|
|
|
"""Compute spectral model fit statistic on a ON OFF Spectrum. |
13
|
|
|
|
14
|
|
|
|
15
|
|
|
Parameters |
16
|
|
|
---------- |
17
|
|
|
model : `~gammapy.spectrum.models.SpectralModel` |
18
|
|
|
Fit model |
19
|
|
|
counts_on : `~gammapy.spectrum.PHACountsSpectrum` |
20
|
|
|
ON Counts spectrum |
21
|
|
|
counts_off : `~gammapy.spectrum.PHACountsSpectrum` |
22
|
|
|
OFF Counts spectrum |
23
|
|
|
livetime : `~astropy.units.Quantity` |
24
|
|
|
Livetime |
25
|
|
|
mask : numpy.array |
26
|
|
|
Mask to apply to the likelihood. |
27
|
|
|
aeff : `~gammapy.irf.EffectiveAreaTable` |
28
|
|
|
Effective area |
29
|
|
|
edisp : `~gammapy.irf.EnergyDispersion` |
30
|
|
|
Energy dispersion |
31
|
|
|
""" |
32
|
|
|
|
33
|
|
|
def __init__( |
34
|
|
|
self, |
35
|
|
|
model=None, |
36
|
|
|
counts_on=None, |
37
|
|
|
counts_off=None, |
38
|
|
|
livetime=None, |
39
|
|
|
mask=None, |
40
|
|
|
aeff=None, |
41
|
|
|
edisp=None, |
42
|
|
|
): |
43
|
|
|
if mask is not None and mask.dtype != np.dtype("bool"): |
44
|
|
|
raise ValueError("mask data must have dtype bool") |
45
|
|
|
|
46
|
|
|
self.counts_on = counts_on |
47
|
|
|
self.counts_off = counts_off |
48
|
|
|
self.livetime = livetime |
49
|
|
|
self.mask = mask |
50
|
|
|
self.aeff = aeff |
51
|
|
|
self.edisp = edisp |
52
|
|
|
|
53
|
|
|
self.model = model |
54
|
|
|
|
55
|
|
|
@property |
56
|
|
|
def alpha(self): |
57
|
|
|
"""Exposure ratio between signal and background regions""" |
58
|
|
|
return self.counts_on.backscal / self.counts_off.backscal |
59
|
|
|
|
60
|
|
|
@property |
61
|
|
|
def model(self): |
62
|
|
|
return self._model |
63
|
|
|
|
64
|
|
|
@model.setter |
65
|
|
|
def model(self, model): |
66
|
|
|
self._model = model |
67
|
|
|
if model is not None: |
68
|
|
|
self._parameters = Parameters(self._model.parameters.parameters) |
69
|
|
|
if self.edisp is None: |
70
|
|
|
self._predictor = SpectrumEvaluator( |
71
|
|
|
model=self.model, |
72
|
|
|
livetime=self.livetime, |
73
|
|
|
aeff=self.aeff, |
74
|
|
|
e_true=self.counts_on.energy.bins, |
75
|
|
|
) |
76
|
|
|
else: |
77
|
|
|
self._predictor = SpectrumEvaluator( |
78
|
|
|
model=self.model, |
79
|
|
|
aeff=self.aeff, |
80
|
|
|
edisp=self.edisp, |
81
|
|
|
livetime=self.livetime, |
82
|
|
|
) |
83
|
|
|
|
84
|
|
|
else: |
85
|
|
|
self._parameters = None |
86
|
|
|
self._predictor = None |
87
|
|
|
|
88
|
|
|
@property |
89
|
|
|
def parameters(self): |
90
|
|
|
if self._parameters is None: |
91
|
|
|
raise AttributeError("No model set for Dataset") |
92
|
|
|
else: |
93
|
|
|
return self._parameters |
94
|
|
|
|
95
|
|
|
@property |
96
|
|
|
def data_shape(self): |
97
|
|
|
"""Shape of the counts data""" |
98
|
|
|
return self.counts_on.data.data.shape |
99
|
|
|
|
100
|
|
|
def npred(self): |
101
|
|
|
"""Returns npred counts vector """ |
102
|
|
|
if self._predictor is None: |
103
|
|
|
raise AttributeError("No model set for this Dataset") |
104
|
|
|
model_npred = self._predictor.compute_npred().data.data |
105
|
|
|
return model_npred |
106
|
|
|
|
107
|
|
|
def likelihood_per_bin(self): |
108
|
|
|
"""Likelihood per bin given the current model parameters""" |
109
|
|
|
on_stat_ = wstat( |
110
|
|
|
n_on=self.counts_on.data.data, |
111
|
|
|
n_off=self.counts_off.data.data, |
112
|
|
|
alpha=self.alpha, |
113
|
|
|
mu_sig=self.npred(), |
114
|
|
|
) |
115
|
|
|
return np.nan_to_num(on_stat_) |
116
|
|
|
|
117
|
|
View Code Duplication |
def likelihood(self, parameters, mask=None): |
|
|
|
|
118
|
|
|
"""Total likelihood given the current model parameters. |
119
|
|
|
|
120
|
|
|
Parameters |
121
|
|
|
---------- |
122
|
|
|
mask : `~numpy.ndarray` |
123
|
|
|
Mask to be combined with the dataset mask. |
124
|
|
|
""" |
125
|
|
|
if self.mask is None and mask is None: |
126
|
|
|
stat = self.likelihood_per_bin() |
127
|
|
|
elif self.mask is None: |
128
|
|
|
stat = self.likelihood_per_bin()[mask] |
129
|
|
|
elif mask is None: |
130
|
|
|
stat = self.likelihood_per_bin()[self.mask] |
131
|
|
|
else: |
132
|
|
|
stat = self.likelihood_per_bin()[mask & self.mask] |
133
|
|
|
|
134
|
|
|
return np.sum(stat, dtype=np.float64) |
135
|
|
|
|
136
|
|
|
@classmethod |
137
|
|
|
def read(cls, filename): |
138
|
|
|
"""Read from file |
139
|
|
|
|
140
|
|
|
For now, filename is assumed to the name of a PHA file where BKG file, ARF, and RMF names |
141
|
|
|
must be set in the PHA header and be present in the same folder |
142
|
|
|
|
143
|
|
|
Parameters |
144
|
|
|
---------- |
145
|
|
|
filename : str |
146
|
|
|
OGIP PHA file to read |
147
|
|
|
""" |
148
|
|
|
observation = SpectrumObservation.read(filename) |
149
|
|
|
return SpectrumDatasetOnOff._from_spectrum_observation(observation) |
150
|
|
|
|
151
|
|
|
# TODO: check if SpectrumObservation is needed in the long run |
152
|
|
|
@classmethod |
153
|
|
|
def _from_spectrum_observation(cls, observation): |
154
|
|
|
"""Creates a SpectrumDatasetOnOff from a SpectrumObservation object""" |
155
|
|
|
|
156
|
|
|
# Build mask from quality vector |
157
|
|
|
quality = observation.on_vector.quality |
158
|
|
|
mask = quality == 0 |
159
|
|
|
|
160
|
|
|
return cls( |
161
|
|
|
counts_on=observation.on_vector, |
162
|
|
|
aeff=observation.aeff, |
163
|
|
|
counts_off=observation.off_vector, |
164
|
|
|
edisp=observation.edisp, |
165
|
|
|
livetime=observation.livetime, |
166
|
|
|
mask=mask, |
167
|
|
|
) |
168
|
|
|
|