New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add some scripts and examples #353

merged 5 commits into from Sep 16, 2015
Jump to file or symbol
Failed to load files and symbols.
+216 −0
Diff settings


Just for now

Viewing a subset of changes. View all

Add examples/

  • Loading branch information...
cdeil committed Sep 16, 2015
commit adf24039a03e81bf43ebcdf3d457c18a8c0dc0ad
Copy path View file
@@ -0,0 +1,216 @@
"""Starting with only gamma-ray event lists, make images and a source catalog.
This is an example showcasing some of the Gammapy features.
With just ~ 100 lines of high-level code we can do this:
- Make a global event list from a datastore (uses `gammapy.obs` and ``)
- Apply an energy and spatial event selection (`EventList` class)
- Bin events into a counts image (uses `` and `gammapy.image`)
- Estimate a background image (uses `gammapy.background`)
- Compute significance = sqrt(TS) image (uses `gammapy.detect`)
- Create a source catalog via a peak finder (uses `gammapy.detect`)
- Make a pretty picture of the images and circle detected sources (uses `gammapy.image`)
You can use this script to run certain steps by commenting in or out the functions in main().
To look at the output significance image and source catalog you can use ds9:
$ ds9 -cmap bb -scale sqrt significance.fits -region sources.reg
- Use adaptive ring (largest sources are cut out)
- The peak finder gives weird results in some regions I don't understand
... need to have a look.
import os
import numpy as np
from import fits
from astropy.wcs import WCS
from astropy.units import Quantity
from astropy.coordinates import Angle, SkyCoord
from astropy.convolution import Gaussian2DKernel
from photutils.detection import find_peaks
from gammapy.obs import DataStore
from import EventListDataset, EventList
from gammapy.background import IterativeKernelBackgroundEstimator, GammaImages
from gammapy.image import binary_disk, binary_ring
from gammapy.detect import compute_ts_map
from gammapy.catalog import to_ds9_region, coordinate_iau_format
HESSFITS_MPP = 'pa/Model_Deconvoluted_Prod26/Mpp_Std/'
REF_IMAGE = 'significance_reference.fits.gz'
TOTAL_EVENTS_FILE = os.path.join(HESSFITS_MPP, 'all_events.fits')
COUNTS_IMAGE = 'counts.fits'
MASK_IMAGE = 'mask.fits'
BACKGROUND_IMAGE = 'background.fits'
TS_IMAGES = 'ts.fits'
SIGNIFICANCE_IMAGE = 'significance.fits'
SOURCE_CATALOG = 'sources.fits'
SOURCE_REGIONS = 'sources.reg'
def make_total_event_list():
"""Make total event list.
TODO: move this function to the datastore class
and the sky box selection to the `make_counts_image` function.
data_store = DataStore(dir=HESSFITS_MPP)
observation_selection = dict(type='sky_box', frame='galactic',
lon=Quantity([-120, 70], 'deg'),
lat=Quantity([-5, 5], 'deg'), border=Quantity(2, 'deg') )
observation_table = data_store.make_observation_table(observation_selection)
# For testing, only process a small subset of observations
# observation_table = observation_table.select_linspace_subset(num=1000)
event_list_files = data_store.make_table_of_files(observation_table, filetypes=['events'])
ds = EventListDataset.vstack_from_files(event_list_files['filename'])
print('Total number of events: {}'.format(len(ds.event_list)))
print('Total number of GTIs: {}'.format(len(ds.good_time_intervals)))
print('Converting EventListDataset to HDUList ...')
hdu_list = ds.to_fits()
print('Writing {}'.format(TOTAL_EVENTS_FILE))
hdu_list.writeto(TOTAL_EVENTS_FILE, clobber=True)
def make_counts_image(energy_band):
"""Apply event selections and bin event positions into a counts image."""
event_list =
n_events = len(event_list)
print('Number of events: {}'.format(n_events))
print('Applying energy band selection: {}'.format(energy_band))
event_list = event_list.select_energy(energy_band)
n_events_selected = len(event_list)
fraction = 100 * n_events_selected / n_events
print('Number of events: {}. Fraction: {:.1f}%'.format(n_events_selected, fraction))
print('Filling counts image ...')
header = fits.getheader(REF_IMAGE)
image = event_list.fill_counts_header(header)
# image = event_lists_to_counts_image(header, TOTAL_EVENTS_FILE)
print('Writing {}'.format(COUNTS_IMAGE))
image.writeto(COUNTS_IMAGE, clobber=True)
def make_background_image():
"""Estimate background image.
See the `IterativeKernelBackgroundEstimator` tutorial and
documentation how it works, or the SciNeGHe 2014 proceeding
by Ellis Owen et al.
radius = Angle(0.2, 'deg')
r_in = Angle(0.3, 'deg')
r_out = Angle(0.7, 'deg')
significance_threshold = 5
mask_dilation_radius = Angle(0.1, 'deg')
max_iterations = 3
hdu =['COUNTS']
binsz = hdu.header['CDELT2']
images = GammaImages(, header=hdu.header)
# TODO: we should have utility functions to initialise
# kernels with angles so that we don't have to convert to pix here.
source_kernel = binary_disk(radius=radius.deg/binsz)
background_kernel = binary_ring(r_in=r_in.deg/binsz,
estimator = IterativeKernelBackgroundEstimator(
images=images, source_kernel=source_kernel, background_kernel=background_kernel,
print('Running background estimation ...')
print('Writing {}'.format(MASK_IMAGE))
estimator.mask_image_hdu.writeto(MASK_IMAGE, clobber=True)
print('Writing {}'.format(BACKGROUND_IMAGE))
estimator.background_image_hdu.writeto(BACKGROUND_IMAGE, clobber=True)
def make_significance_image():
"""Make significance = sqrt(TS) image using a Gaussian kernel.
gauss_kernel_sigma = 5 # pix
header = fits.getheader(REF_IMAGE)
counts = fits.getdata(COUNTS_IMAGE)
background = fits.getdata(BACKGROUND_IMAGE)
exposure = 1e11 * np.ones_like(counts, dtype='float32')
kernel = Gaussian2DKernel(gauss_kernel_sigma)
print('Computing TS image ...')
result = compute_ts_map(counts, background, exposure, kernel)
print('TS map computation took: {}'.format(result.runtime))
print('Writing {}'.format(TS_IMAGES))
result.write(TS_IMAGES, header=header, overwrite=True)
def make_source_catalog():
"""Make source catalog from images.
TODO: use other images to do measurements for the sources,
e.g. excess, npred, flux.
significance_threshold = 7
hdu =['sqrt_ts']
header = fits.getheader(REF_IMAGE)
# import IPython; IPython.embed()
wcs = WCS(header)
print('Running find_peaks ...')
table = find_peaks(, threshold=significance_threshold, wcs=wcs)
print('Number of sources detected: {}'.format(len(table)))
# Add some useful columns
icrs = SkyCoord(table['icrs_ra_peak'], table['icrs_dec_peak'], unit='deg')
galactic = icrs.galactic
table['Source_Name'] = coordinate_iau_format(icrs, ra_digits=5, prefix='J')
table['GLON'] = galactic.l.deg
table['GLAT'] = galactic.b.deg
# table.show_in_browser(jsviewer=True)
print('Writing {}'.format(SOURCE_CATALOG))
table.write(SOURCE_CATALOG, overwrite=True)
ds9_string = to_ds9_region(table, label='Source_Name')
print('Writing {}'.format(SOURCE_REGIONS))
with open(SOURCE_REGIONS, 'w') as fh:
# TODO: move this to `make_significance_image`.
# At the moment the TS map output images don't have WCS info in the header
hdu = fits.PrimaryHDU(, header=header)
print('Writing {}'.format(SIGNIFICANCE_IMAGE))
hdu.writeto(SIGNIFICANCE_IMAGE, clobber=True)
def main():
"""Run the whole analysis chain.
Here the steps communicate via FITS files.
energy_band = Quantity([1, 10], 'TeV')
if __name__ == '__main__':
ProTip! Use n and p to navigate between commits in a pull request.