Fermi GBM Classification Algorithm (gdt.missions.fermi.gbm.classification)

The Fermi GBM has an on-board classifier for high-energy transients detected by the flight software. This classifier is operated in real-time on the spacecraft and it uses information such as the on-board localization and hardness ratio to determine if the trigger is astrophysical, and if so, what is the likely source of the trigger.

Note

This GDT implementation of the GBM classification algorithm is a close approximation of the algorithm operating on the spacecraft; however there may be small differences compared to the exact flight software algorithm.

While there are a couple of ways to employ this algorithm, we will use the method that reads the required information from a trigdat (trigger data) file (see this documentation for more info on working with trigdat). The alternative method requires cobbling together the requisite inputs, which an advanced user may wish to do when finding a potential signal in continuous GBM that did not produce an on-board trigger (see the classify() method for more info about the inputs).

To start, we will create and initialize our Classification object:

>>> from gdt.missions.fermi.gbm.classification import Classification
>>> classifier = Classification()

Next, we open a trigdat file:

>>> from gdt.core import data_path
>>> from gdt.missions.fermi.gbm.trigdat import Trigdat
>>> filepath = data_path.joinpath('fermi-gbm/glg_trigdat_all_bn170101116_v01.fit')
>>> trigdat = Trigdat.open(filepath)

Now we can pass the Trigdat object to the Classification object, which performs an approximation of the source classification routine used on-board GBM. During this step we apply the following settings:

• use_loc=True performs the classification using the detectors that are closest to the source location. When False the routine will instead sort the detectors by the largest count rates in the 50-300 keV energy band.

• verbose=True prints the trigdat information used as part of the classification algorithm. Setting this to False` or ignoring this argument will silence this output.

  >>> det_nums = classifier.classify_trigdat(trigdat, use_loc=True, verbose=True)
  Trigger Algorithm = 6
  Location = (RA 59.167, Dec 9.333) deg
  Location Err = 5.317 deg
  Rates
   [[ 64. 279. 223. 246. 269.  46.  38.  72.]
   [ 54. 283. 219. 228. 235.  40.  43.  57.]
   [ 42. 202. 197. 194. 202.  40.  56.  37.]
   [ 67. 290. 218. 191. 184.  36.  36.  74.]
   [ 43. 216. 202. 204. 176.  39.  67.  47.]
   [ 35. 154. 183. 198. 205.  45.  90.  29.]
   [ 82. 364. 269. 243. 265.  48.  65.  20.]
   [ 99. 410. 291. 245. 253.  45.  44.  46.]
   [ 70. 392. 236. 211. 195.  39.  62.  36.]
   [ 82. 518. 379. 375. 342.  51.  61.  16.]
   [ 91. 483. 396. 371. 348.  46.  68.  22.]
   [ 74. 438. 303. 307. 279.  49.  32.  47.]
   [623. 399. 525. 224.  36.  30.  21.  95.]
   [702. 423. 423. 172.  28.  23.  21.  68.]]
  Background
   [[ 63. 276. 195. 145. 153.  39.  33.  65.]
   [ 56. 277. 205. 148. 147.  40.  43.  54.]
   [ 45. 206. 177. 145. 156.  38.  64.  41.]
   [ 71. 303. 210. 160. 159.  42.  40.  82.]
   [ 45. 207. 181. 149. 160.  40.  76.  38.]
   [ 38. 145. 155. 138. 172.  47.  98.  27.]
   [ 72. 325. 201. 157. 141.  34.  69.  21.]
   [ 88. 387. 240. 161. 140.  32.  45.  45.]
   [ 68. 372. 232. 163. 141.  34.  61.  33.]
   [ 70. 403. 253. 167. 137.  31.  61.  16.]
   [ 86. 382. 257. 169. 136.  31.  59.  19.]
   [ 63. 349. 227. 157. 143.  35.  31.  52.]
   [550. 402. 488. 240.  40.  29.  22.  91.]
   [412. 347. 387. 168.  31.  24.  18.  75.]]
  

The classify_trigdat() method returns a list with the detector numbers for the top 2 detectors closest to the source location. Let’s check which ones they are:

>>> from gdt.missions.fermi.gbm.detectors import GbmDetectors
>>> print("\nClosest detectors:")
>>> for num in det_nums:
>>>     print("  {}".format(GbmDetectors.from_num(num).name))
Closest detectors:
  n9
  na

We can also inspect the top two source classifications, ranked according to probability:

>>> print("\nTop 2 classifications (this calculation):")
>>> rank = classifier.rankings()[:2]
>>> for r in classifier.rankings()[:2]:
>>>     print("  prob = {:.2f} {}".format(r[1], r[0].name))
Top 2 classifications (this calculation):
  prob = 0.86 GRB
  prob = 0.10 GROJ_0422_32

For comparison, let’s also show the classifications determined by the flight software on-board the spacecraft:

>>> fsw_loc = trigdat.fsw_locations[-1]
>>> print("\nTop 2 classifications (flight):")
>>> for c, r in [fsw_loc.top_classification, fsw_loc.next_classification]:
>>>     print("  prob = {:.2f} {}".format(r, c))
Top 2 classifications (flight):
  prob = 0.86 GRB
  prob = 0.09 GROJ422

And since we are curious, we can see the full breakdown of classification probabilities:

>>> print(classifier)
Classification(
               ERROR               0.0000
               UNRELIABLE_LOCATION 0.0000
               LOCAL_PARTICLES     0.0000
               BELOW_HORIZON       0.0000
               GRB                 0.8633
               GENERIC_SGR         0.0044
               GENERIC_TRANSIENT   0.0359
               DISTANT_PARTICLES   0.0000
               SOLAR_FLARE         0.0000
               CYG_X1              0.0000
               SGR_1806_20         0.0000
               GROJ_0422_32        0.0964
               RESERVED_1          0.0000
               RESERVED_2          0.0000
               RESERVED_3          0.0000
               RESERVED_4          0.0000
               RESERVED_5          0.0000
               RESERVED_6          0.0000
               RESERVED_7          0.0000
               TGF                 0.0000)

Reference/API

gdt.missions.fermi.gbm.classification Module

Classes

Classification()	Class for calculating source classification probabilities
TriggerClasses(value[, names, module, ...])	Definition of numeric trigger classes

Class Inheritance Diagram