I came across this simplexy source detection method:
https://github.com/astrojs/simplexy.js
https://github.com/astrojs/simplexy.js/blob/master/src/simplexy.coffee
http://trac.astrometry.net/wiki/SimplexyTuneup
http://trac.astrometry.net/browser/trunk/src/astrometry/blind/simplexy.c
https://www.stsci.edu/trac/ssb/stsci_python/browser/GoogleSky/trunk/astrometry.net/quads/simplexy.py
http://astrometry.net/svn/trunk/src/astrometry/blind/image2xy.py

@adonath Seems like it's a very simple and popular detection method ... maybe you or I should implement it in gammapy.detect or photutils.
I couldn't find a method paper ... I guess it's just defined by this description and the original C implementation:
http://trac.astrometry.net/browser/trunk/src/astrometry/blind/simplexy.c#L42

Lafferty & Wyatt (1995) gives a fancy method to position points within a bin, which is commonly used in TeV gamma-ray astronomy to compute flux points.

See e.g. Equation 3.7 on page 63 and the footnote on page 65 in Hoppe.

This should be implemented in gammapy/spectrum/flux_point.py

Issue #61 should be done first.

Include this and add a Python wrapper:
https://github.com/cdeil/root_poisson_stats

Implement some book-keeping classes:

RunInfo:
- livetime
- RA, DEC
- mean_zenith

and a RunList class which is a list of RunInfos.

Both should have read / write e.g. in JSON or ??? format.

There is already a gammapy.obs module, but it's currently almost empty. See here and here.

I should update the gammapy .travis.yml file to fix this:
astropy/astropy@513b002#commitcomment-4148009

And maybe start using wheels to speed up the tests ...

Implement image_to_table and table_to_image conversion functions.
They should take a mask of pixels of interest and optionally give pixel and / or world coordinate columns.
The WCS info will be lost in the table of course.

@adonath Could you please fix the following error?

_______________________________________________________________________________ test_SNR_Truelove ________________________________________________________________________________

    def test_SNR_Truelove():
        snr = SNR_Truelove()
        assert_approx_equal(snr.L(1e3), 1.0768456645602824e+33)
>       assert_approx_equal(snr.r_out(1e3), 42)

tevpy/astro/tests/test_snr.py:14: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <tevpy.astro.snr.SNR_Truelove object at 0x10bedb8d0>, t = 31557600000.000004

    def r_out(self, t):
        """Outer shell radius (pc) at age t (yr).
            """
        t = YEAR_TO_SEC * t
        # Sedov Taylor Phase
        t_ST = 0.52 * self.t_c
>       R_FE = self.r_reverse_free_expansion(t_ST)
E       AttributeError: 'SNR_Truelove' object has no attribute 'r_reverse_free_expansion'

tevpy/astro/snr.py:166: AttributeError

The Maps container class constructor logic is broken at the moment (doesn't show up on travis-ci for some reason):
https://gist.github.com/cdeil/7869324

It should be re-written, probably use gammapy.stats.Stats and have FITS I/O in methods from_fits and to_fits instead of sub-classing HDUList.

The help output from the command line tools ("scripts") as well as examples should be included in the docs for end-users that don't want to use Python or know how it works.
#230 makes this possible ... the script functions (but not their command line tool help output) are now listed here: https://gammapy.readthedocs.org/en/latest/scripts/index.html

We still have to programatically get at the help output and integrate this in the Sphinx build somehow ...

We could make a movie (e.g. using Blender or a JavaScript 3D animation like this to illustrate how the Milky way looks for different source population models and instrument sensitivities / resolutions.

cc @adonath

This is just a reminder ... it's an astropy issue:
astropy/astropy#1256

I'll disable Python 3.2 tests on travis-ci for now in gammapy until this gets fixed.

The current GeneralRandom class should be improved (work with n-dim arrays, rename, document, unit tests).

I think this can be implemented better with np.choice when passing a=pdf and p=pdf.

It would also be nice to add functions that consider the array a discretised version of a continuous function and then sample that continuous function, similar to what is available in ROOT:
http://root.cern.ch/root/html/src/TF2.cxx.html#QsZi6C

I've made a feature request for this in scipy: scipy/scipy#3747

Implement some book-keeping classes:

RunInfo:
- livetime
- RA, DEC
- mean_zenith

and a RunList class which is a list of RunInfos.

Both should have read / write e.g. in JSON format.

In Gammapy we should support existing spectrum file formats.
Also we should make it easy to use other existing tools.

These might be useful for visualisation or even analysis, there might be more:

• http://tools.asdc.asi.it/SED/
• http://www.asdc.asi.it/tgevcat/
• http://cxc.cfa.harvard.edu/iris/latest/index.html
• https://github.com/jbudynk/sedstacker

I think these all work on spectra = flux points, I'm not sure if any know about effective areas or other IRFs and counts spectra.

And there's of course the X-ray PHA, ARF, RMF standards and tools like Sherpa, Fermi Science tools and Gammalib and Naima that support those or other slightly different formats.

Overall it's pretty difficult to decide what exactly should go in gammapy.spectrum and what is needed is both code to handle the most useful formats, as well as high-level documentation which formats we support (e.g. write this page and which other tools can be useful for SED modeling / analysis / fitting.

@joleroi, @ignasi-reichardt, @registerrier – Maybe we can have a telcon and hash out a proposal what exactly we want next week?

At the moment test_image fails like this:
https://gist.github.com/cdeil/8936875

I don't want to look into this right now, so I'll just disable this test.

I am trying to add Cython files to this package which I forked from astropy/package-template, but for some reason setup.py doesn't find cython:

$ python setup.py build
running build
running build_py
running build_ext
error: tevpy/example_c.c: Could not find C file tevpy/example_c.c for Cython file tevpy/example_c.pyx when building extension tevpy.example_c. Cython must be installed to build from a git checkout

To reproduce:

git clone https://github.com/gammapy/tevpy.git
cd tevpy
curl -L https://raw.github.com/astropy/package-template/master/packagename/example_c.pyx > tevpy/example_c.pyx
python setup.py build

@astrofrog @iguananaut I don't understand why this works for other affiliated packages (I tried with package-template and astroquery), but not in tevpy! Can one of you please have a look?

Here's an old script I wrote to work with GALPROP output cubes:
https://github.com/gammapy/gammapy/blob/master/examples/galprop.py

@ellisowen Let's discuss how to best implement this functionality and then you re-write it and add documentation and tests.

We should add basic data and methods for TeV telescope observer calculations.

E.g. TeVCat has for each TeV telescope location and TeV target the "Vis-plot this source" and "vis-plot full year" plots ... see e.g. the Crab as an example.

Since astropy.coordinates doesn't contain Sky-to-horizon coordinate transformations yet, we can use PyEphem as shown in this example to do the computations.

Other examples worth looking at are the Python observer or visibility-map packages and observation apps like iobserve.

Reproducing the TeVCat plots mentioned above should be easy as a start and then we can ask for user feature requests.

Suggestion by @eteq in astropy/astropy.github.com#3 (comment) : gammapy.spectrum and gammapy.spectral_cube should use specutils classes (of course this isn't very yet, so this is an "in the future" suggestion. I'm not sure if you know about this idea, but the eventual plan is to build a base spectral class that can then be subclassed by something like gammapy to have convinentions appropriate for the gamma ray subfield)

We should implement the reflected background method.

It's not clear to me yet what a good API (class or function?) and implementation (cartesian FOV or spherical sky coordinate system?) for this is.

Some code that computes reflected regions is available already in gammapy/background/reflected.py

It would probably be best to start with a concrete use case, e.g. compute n_on and n_off spectra for XSPEC-style analysis using eventlists and exclusion regions as input.

Currently there's the ts_image function and TSMapCalculator in gammapy.detect.

We should probably also add a command line tool to make TS maps ... here's a very old script as an example: https://gist.github.com/cdeil/58ced17e44d9e8946ea8

All functions working with pixel coordinates need to be reviewed to make sure they follow the astropy pixel coordinate convention:
http://photutils.readthedocs.org/en/latest/photutils/index.html#coordinate-convention-in-photutils

Specifically some functions currently contain a shift argument with default 0.5 ... I think this can be removed and simply follow this convention.

Suggestion by @eteq from http://github.com/astropy/astropy.github.com/pull/3#issuecomment-35200275 : The source/population models would be natural use cases for astropy.modeling, as well as the PSF models.

All "edit on github" links in the docs are broken.

E.g. the top-level page at the moment points to this non-existing URL:
https://github.com/gammapy/gammapy/tree/v0.1/docs/index.rst

After #68 is merged, we should add an XSPEC example to gammapy.data and example plots of the response and analysis as a tutorial.

A Fermi observation of a GRB would be nice:

• http://fermi.gsfc.nasa.gov/ssc/data/analysis/scitools/help/gtrspgen.txt
• http://fermi.gsfc.nasa.gov/ssc/data/analysis/scitools/lat_grb_analysis.html
• http://fermi.gsfc.nasa.gov/ssc/data/analysis/documentation/Cicerone/Cicerone_Likelihood/XSPEC_for_Fermi.html

The Fermi tools generate RSP files (see here which are the produce of the ARF and RMF. We need a gammapy reader or class for this.

I've had problems getting accurate results when numerically integrating power-law type functions (spectral models like curved or exponential cutoff power law) using scipy.quad or other integrators.

I think using a log(x) spacing instead of x and using a simple integration method that takes vectors of x and y as input like cumtrapz or scipy.simps
should give good results.

Also check all GammaLib results for flux and eflux and compare speed / accuracy. Usage example:

from gammalib import GModelSpectralPlaw, GModelSpectralPlaw2, GModelSpectralExpPlaw, GModelSpectralLogParabola
from gammalib import GEnergy, GTime
model = GModelSpectralPlaw(1, 2, GEnergy(1, 'TeV'))
print model.eval(gammalib.GEnergy(10, 'TeV'), gammalib.GTime(1, 's'))
print model.flux(GEnergy(1, 'TeV'), GEnergy(10, 'TeV'))
print model.eflux(GEnergy(1, 'TeV'), GEnergy(10, 'TeV'))

Links:

• https://cta-redmine.irap.omp.eu/projects/gammalib/wiki/How_to_integrate_a_function

Implement the xspec tool.
See docstring for what it should do.

Since gammapy will never be integrated in astropy core, I think it makes sense to move all docs from docs/gammapy to docs so that they will show up at
https://gammapy.readthedocs.org/en/latest/
instead of at
https://gammapy.readthedocs.org/en/latest/gammapy/index.html
where they are now.

TestMaps.test_make_derived_maps currently fails:
https://gist.github.com/cdeil/8936825

For now I'll just disable the test because I don't have time to fix it now.

There should be a tool to simulate background event lists to generate test data sets.

What I have in mind is a simple analytical formula

background_rate(zenith, offset, mscw)

that gives the background rate in units counts / second / deg ^ 2 for a given zenith angle, offset in the FOV and mean scaled width mscw parameter.

The point is not to have a very realistic model, just any model so that all the background methods can be tested.

Proposal:

• total rate (for all offsets, mscw) is a simple polynomial in zenith.
• FOV distribution is 2D Gauss and mscw distribution 1D Gauss with varying widths (how?)

The gammapy.spectrum sub-package contains a lot of old code that needs to be reviewed.
Keep the useful parts and add unit tests / docs.
Remove the superfluous stuff ...

I also have this super-old script for PWN spectra which can probably be thrown away except for this analytical formula for the exponential-cutoff power-law spectrum

The docs PDF build gives errors:
https://readthedocs.org/builds/gammapy/

and doesn't look good:
https://media.readthedocs.org/pdf/gammapy/latest/gammapy.pdf

First check package-template if the problem is there already.

I'm using astropy.utils.data.get_pkg_data_filename in a few places in tevpy (see here).

This causes four tevpy Sphinx build warnings:

building [html]: all source files
updating environment: 126 added, 0 changed, 0 removed
reading sources... [100%] tevpy/utils/index                                    rocity
/Users/deil/code/tevpy/docs/tevpy/detect/index.rst:85: WARNING: toctree references unknown document u'_generated/astropy.utils.data.get_pkg_data_filename'
looking for now-outdated files... none found
pickling environment... done
checking consistency... done
preparing documents... done
writing output... [100%] tevpy/utils/index                                     ocity
None:None: WARNING: toctree contains reference to nonexisting document '_generated/astropy.utils.data.get_pkg_data_filename'
<autosummary>:None: WARNING: toctree contains reference to nonexisting document '_generated/astropy.utils.data.get_pkg_data_filename'
None:None: WARNING: toctree contains reference to nonexisting document '_generated/astropy.utils.data.get_pkg_data_filename'
writing additional files... (31 module code pages) _modules/index
 genindex py-modindex np-modindex search
copying static files... done
dumping search index... done
dumping object inventory... done
build succeeded, 4 warnings.

It's annoying because it makes the travis-ci sphinx build fail:
https://travis-ci.org/gammapy/tevpy/jobs/9869684#L686

@eteq @astrofrog @iguananaut Do you know why this is happening? How can I fix it?

To showcase the different parts of astropy and gammapy we should make simulated gamma-ray images of the Milky Way.

The tevpy docs build now fails because of a problem with astropy.wcs (see below and here.
I don't see this error on my local machine.

@mdboom Do you know what the problem is? Is this a problem in astropy or am I doing something wrong?

Obtaining astropy from git+http://github.com/astropy/astropy.git#egg=astropy (from -r docs/rtd-pip-requirements.txt (line 4))
  Updating /home/docs/checkouts/readthedocs.org/user_builds/tevpy/envs/latest/src/astropy clone
  Running setup.py egg_info for package astropy
    Traceback (most recent call last):
      File "<string>", line 16, in <module>
      File "/home/docs/checkouts/readthedocs.org/user_builds/tevpy/envs/latest/src/astropy/setup.py", line 82, in <module>
        package_dirs)
      File "astropy/setup_helpers.py", line 1028, in update_package_files
        extensions.extend(setuppkg.get_extensions())
      File "astropy/wcs/setup_package.py", line 206, in get_extensions
        generate_c_docstrings()
      File "astropy/wcs/setup_package.py", line 122, in generate_c_docstrings
        from astropy.wcs import docstrings
      File "astropy/wcs/__init__.py", line 27, in <module>
        from .wcs import *
      File "astropy/wcs/wcs.py", line 82, in <module>
        WcsError = _wcs.WcsError
    AttributeError: 'module' object has no attribute 'WcsError'
    Complete output from command python setup.py egg_info:
    Traceback (most recent call last):

  File "<string>", line 16, in <module>

  File "/home/docs/checkouts/readthedocs.org/user_builds/tevpy/envs/latest/src/astropy/setup.py", line 82, in <module>

    package_dirs)

  File "astropy/setup_helpers.py", line 1028, in update_package_files

    extensions.extend(setuppkg.get_extensions())

  File "astropy/wcs/setup_package.py", line 206, in get_extensions

    generate_c_docstrings()

  File "astropy/wcs/setup_package.py", line 122, in generate_c_docstrings

    from astropy.wcs import docstrings

  File "astropy/wcs/__init__.py", line 27, in <module>

    from .wcs import *

  File "astropy/wcs/wcs.py", line 82, in <module>

    WcsError = _wcs.WcsError

AttributeError: 'module' object has no attribute 'WcsError'

----------------------------------------
Command python setup.py egg_info failed with error code 1 in /home/docs/checkouts/readthedocs.org/user_builds/tevpy/envs/latest/src/astropy

We should add a script gp-morph-fit-sherpa that allows to perform morphology fits with Sherpa using an on-off likelihood or on-likelihood fit (using background = alpha * n_off).

Here's an old version of mine I have to integrate into gp that does the on-likelihood fit:
https://gist.github.com/cdeil/5974689

@registerrier Can you make a pull request with your script that does the on-off likelihood morphology fit with wstat?

Some background modelling is sometimes done in the FOV and then projected on a spherical coordinate system at the end.
Exclusion regions are often defined on the sky and need to be projected to the FOV system to be used there.
Implement a method to project back and forth (TAN WCS projection).

• https://github.com/gammapy/gammapy/blob/master/examples/add_gnomonic_projection.py
• https://github.com/gammapy/gammapy/blob/master/examples/horizon_coord_transformations.py
  Needs to be moved to gammapy.utils.coordinates, cleaned up, unit tests added, docs added.

Some references ... I plan to work on this next week:

• https://github.com/astrofrog/reproject/pull/60/files
• https://heasarc.gsfc.nasa.gov/docs/heasarc/ofwg/docs/events/ogip_94_003/ogip_94_003.html#127
• https://github.com/cta-observatory/ctapipe/blob/master/ctapipe/coordinates/frames.py

Check against Gammalib / ctools:

• http://cta.irap.omp.eu/gammalib-devel/doxygen/classGCTABackground3D.html#1442f55d3b958a8d7a1e9160665ad0c2
• http://gammalib.sourceforge.net/doxygen/classGCTAPointing.html#b093e7b0bc07520669f05a578490d44c
• http://gammalib.sourceforge.net/doxygen/classGCTAPointing.html#5485315bf59fdc1c4d164d059129ea24

Here's an example showing the problem:

/home/eowen/.local/lib/python2.7/site-packages/gammapy-0.1-py2.7.egg/gammapy/image/utils.pyc in _lookup_world(image, lon, lat)
    136     wcs = WCS(image.header)
    137     x, y = wcs.wcs_world2pix(lon, lat, 0)
--> 138     return _lookup_pix(image.data, x, y)
    139 
    140 

/home/eowen/.local/lib/python2.7/site-packages/gammapy-0.1-py2.7.egg/gammapy/image/utils.pyc in _lookup_pix(image, x, y)
    124     # Return it's value

    125     # Note that numpy has index order (y, x)

--> 126     values = image[y_int, x_int]
    127     return values
    128 

IndexError: index (9400) out of range (0<=index<9399) in dimension 1

This method is describe here (HESS ParisAnalysis software is not publicly available):
http://llr.in2p3.fr/sites/hess/hess-documentation/parisanalysis/

The numerical integration to compute integral fluxes doesn't seem to work well ... I had this note:

    # @todo there are integration problems with e2=1e6.
    # test and use the integrator that works in log space!!!
    """
    In [36]: spec.crab.int_flux(0.2, 1e4, reference='hess_ecpl')
[  2.43196827e-10] [  2.61476507e-18]
Out[36]: array([  2.43196827e-10])

In [37]: spec.crab.int_flux(0.2, 1e5, reference='hess_ecpl')
Warning: The algorithm does not converge.  Roundoff error is detected
  in the extrapolation table.  It is assumed that the requested tolerance
  cannot be achieved, and that the returned result (if full_output = 1) is
  the best which can be obtained.
[  2.43283459e-10] [  4.37319063e-10]
Out[37]: array([  2.43283459e-10])

In [38]: spec.crab.int_flux(0.2, 1e6, reference='hess_ecpl')
[  6.40098358e-48] [  1.27271100e-47]
Out[38]: array([  6.40098358e-48])
    """

• Make the integration method robust
• Add several unit tests

The looping over array entries and the iterative optimisation for each entry in gammapy.stats.sensitivity_on_off should be implemented in Cython to get good speed.

Before doing that unit tests and a benchmark should be added.

Implement an adaptive ring background method in gammapy/background/ring.py.

The ring background method is described in Berge 2007, the adaptive ring background method in Carrigan 2013.

@astrofrog @eteq I can't figure out what I did wrong with the astropy affiliated package template so that now python setup.py install doesn't install any Python files:

warning: install_lib: 'build/lib' does not exist -- no Python modules to install

Indeed build/lib doesn't exist, it's called build/lib.macosx-10.7-x86_64-2.7.
Full log here: https://gist.github.com/cdeil/5586877

FYI: I am using the astropy affiliated package template for this project just to get the build / test / docs infrastructure, I most likely won't develop this into an astropy affiliated package, it's just a prototype for a proper implementation in the C++ library gammalib.

Sometimes HESS makes images in FITS or spectra in ascii available for download here (search for "aux"):

• http://www.mpi-hd.mpg.de/hfm/HESS/pages/publications/pubs_all.shtml

We can use these as test datasets ...

• http://www.mpi-hd.mpg.de/hfm/HESS/pages/publications/auxiliary/HESSVelaX_fluxmap.fits

In the poisson_stats_image example possibly both during simulation and during fitting the model is not integrated over bins?

Implement a FOV background method (see Berge 2007) in gammapy/background/fov.py.

The main thing to implement is computing the 2D image from 1D radial acceptance curves, and then to scale to events outside exclusion regions and apply a FOV offset cut.

• Should the astropy logo replaced with the gammapy logo?
• The current favicon should be replaced with this one.

See https://gammapy.readthedocs.org/en/latest/index.html

There is some old code in pyfact.py to do template background estimation as in Rowell 2003.

This should be cleaned up (nice API, add docs, add tests) in gammapy.background.

Add a utility function that computes a source name string in these formats:

• HESS J0534+220
• 2FGL JHHMM.m+DDMM, constructed according to IAU Specifications for Nomenclature; m is decimal minutes of R.A.; in the name R.A. and decl. are truncated at 0.1 decimal minutes and 1, respectively

This should probably be in astropy ... check first if it already is or make PR there.