Command line tools (gammapy.scripts)


The Gammapy command line interface (CLI) described here is experimental and only supports a small sub-set of the functionality available via the Gammapy Python package. We have added a few sections at the bottom of this page to explain the current Implementation, Limitations and Plan. And since we don’t offer much here yet, at least we describe how you can Write your own CLI.


Currently, Gammapy is first and foremost a Python package. This means that to use it you have to write a Python script or Jupyter notebook, where you import the functions and classes needed for a given analysis, and then call them, passing parameters to configure the analysis.

That said, for some very commonly used and easy to configure analysis tasks we have implemented a command line interface (CLI). It is automatically installed together with the Gammapy python package.


To execute the Gammapy CLI, type the command gammapy at your terminal shell (not in Python):

$ gammapy --help

or equivalently, just type this:

$ gammapy

Either way, the command should print some help text to the console and then exit:

$ gammapy

  Gammapy command line interface.

  Gammapy is a Python package for gamma-ray astronomy.

  For further information, see

  --log-level [debug|info|warning|error]
                                  Logging verbosity level
  --ignore-warnings               Ignore warnings?
  --version                       Print version and exit
  -h, --help                      Show this message and exit.

  check  Run checks for Gammapy
  image  Analysis - 2D images
  info   Display information about Gammapy

All CLI functionality for Gammapy is implemented as sub-commands of the main gammapy command. If a command has sub-commands, they are listed in the help output. E.g. the help output from gammapy above shows that there is a sub-command called gammapy image. Actually, gammapy image itself isn’t a command that does something, but another command group that is used to group sub-commands:

$ gammapy image
Usage: gammapy image [OPTIONS] COMMAND [ARGS]...

  Analysis - 2D images

  -h, --help  Show this message and exit.

  bin  Bin events into an image.
  fit  Fit morphology model to image using Sherpa.
  ts   Compute TS image.

Finally, gammapy image bin is a proper sub-sub-command that does something, it doesn’t have any sub-commands itself, just arguments and options. If you call it without passing the required arguments, you will get an error:

$ gammapy image bin

Error: Missing argument "event_file".

Use --help to see the help text and available options:

$ gammapy image bin --help

  Bin events into an image.

  You have to give the event, reference and out FITS filename.

  --overwrite  Overwrite existing files?
  -h, --help   Show this message and exit.

So now you know how the Gammapy CLI is structured and how to discover all available sub-commands, arguments and options.

Command not found

Usually tools that install Gammapy (e.g. setuptools via python install or pip or package managers like conda) will put the gammapy command line tool in a directory that is on your PATH, and if you type gammapy the command is found and executed.

However, due to the large number of supported systems (Linux, Mac OS, Windows) and different ways to install Python packages like Gammapy (e.g. system install, user install, virtual environments, conda environments) and environments to launch command line tools like gammapy (e.g. bash, csh, Windows command prompt, Jupyter, …) it is not unheard of that users have trouble running gammapy after installing it.

This usually looks like this:

$ gammapy
-bash: gammapy: command not found

If you just installed Gammapy, search the install log for the message “Installing gammapy script” to see where gammapy was installed, and check that this location is on your PATH:

echo $PATH

If you don’t manage to figure out where the gammapy command line tool is installed, you can try calling it like this instead:

$ python -m gammapy

This also has the advantage that it avoids issues where users have multiple versions of Python and Gammapy installed and accidentally launch one they don’t want because it comes first on their PATH. For the same reason these days the recommended way to use e.g. pip is via python -m pip.

If this still doesn’t work, check if you are using the right Python and have Gammapy installed:

$ which python
$ python -c 'import gammapy'

To see more information about your shell environment, these commands might be helpful:

$ python -m site
$ python -m gammapy info
$ echo $PATH
$ conda info -a # if you're using conda

If you’re still stuck or have any question, feel free to ask for help with installation issues on the Gammapy mailing list of Slack any time!


Here’s one example what you can do with the Gammapy CLI: use the gammapy image bin command to create a counts image from a Fermi-LAT event list as well as a FITS image that serves as a reference geometry (projection, center, binning) for the counts image we’re creating.

$ gammapy image bin --help

  Bin events into an image.

  You have to give the event, reference and out FITS filename.

  --overwrite  Overwrite existing files?
  -h, --help   Show this message and exit.

$ gammapy image bin \
    $GAMMAPY_EXTRA/datasets/fermi_2fhl/2fhl_events.fits.gz \
    $GAMMAPY_EXTRA/datasets/fermi_survey/all.fits.gz \
INFO:gammapy.scripts.image_bin:Executing cli_image_bin
INFO:gammapy.scripts.image_bin:Reading /Users/deil/code/gammapy-extra/datasets/fermi_2fhl/2fhl_events.fits.gz
INFO:gammapy.scripts.image_bin:Reading /Users/deil/code/gammapy-extra/datasets/fermi_survey/all.fits.gz
INFO:gammapy.scripts.image_bin:Writing out.fits

If you have the FTOOLS installed or other tools that can work with the files that Gammapy supports, you can of course use them together:

$ ftlist $GAMMAPY_EXTRA/datasets/fermi_2fhl/2fhl_events.fits.gz H

        Name               Type       Dimensions
        ----               ----       ----------
HDU 1   Primary Array      Null Array
HDU 2   EVENTS             BinTable    23 cols x 60978 rows
HDU 3   GTI                BinTable     2 cols x 36589 rows

$ ftlist out.fits H

        Name               Type       Dimensions
        ----               ----       ----------
HDU 1   COUNTS             Image      Real8(2001x101)

$ ds9 out.fits


Here is auto-generated documentation for all available sub-commands, arguments and options of the gammapy command line interface (CLI).

It’s not very readable at the moment. With the current formatting it’s a bit hard to tell where documentation for a new sub-command starts and what level of subcommand one is looking at for a given heading. Maybe change to one page per sub-command?


Gammapy command line interface (CLI).

Gammapy is a Python package for gamma-ray astronomy.

Use --help to see available sub-commands, as well as the available arguments and options for each sub-command.

For further information, see and

 Examples ——–

 $ gammapy –help $ gammapy –version $ gammapy info –help $ gammapy info



--log-level <log_level>

Logging verbosity level


Ignore warnings?


Print version and exit


Run checks for Gammapy

gammapy check [OPTIONS] COMMAND [ARGS]...

Check logging

gammapy check logging [OPTIONS]

Run Gammapy tests

gammapy check runtests [OPTIONS]


Analysis - 2D images

gammapy image [OPTIONS] COMMAND [ARGS]...

Bin events into an image.

You have to give the event, reference and out FITS filename.




Overwrite existing files?



Required argument


Required argument


Required argument


Fit morphology model to image using Sherpa.

Uses initial parameters from a JSON file (for now only Gaussians).

gammapy image fit [OPTIONS]


--counts <counts>

Counts FITS file name

--exposure <exposure>

Exposure FITS file name

--background <background>

Background FITS file name

--psf <psf>

PSF JSON file name

--sources <sources>

Sources JSON file name (contains start values for fit of Gaussians)

--roi <roi>

Region of interest (ROI) file name (ds9 reg format)

--outfile <outfile>

Output JSON file with fit results


Compute TS image.

The INPUT_FILE and OUTPUT_FILE arguments are FITS filenames.

The INPUT_FILE FITS file must contain the following HDU extensions:  * ‘counts’ – Counts image * ‘background’ – Background image * ‘exposure’ – Exposure image



--psf <psf>

JSON file containing PSF information.

--morphology <morphology>

Which source morphology to use for TS calculation.

--width <width>

Width of the shell, measured as fraction of the inner radius.

--scales <scales>

Angular kernel scales in deg (comma-separated list).

--downsample <downsample>

Downsample factor of the data to obtain optimal performance. Must be power of 2. Can be “auto” to choose the downsample factor automatically depending on the scale.


Whether to compute a residual TS image. If a residual TS image is computed an excess model has to be provided using the “–model” parameter.

--model <model>

Input excess model FITS file name

--threshold <threshold>

Minimal required initial (before fitting) TS value, that the fit is done at all.


Overwrite output files.



Required argument


Required argument


Display information about Gammapy

gammapy info [OPTIONS]


--system, --no-system

Show system info

--version, --no-version

Show version info

--dependencies, --no-dependencies

Show dependencies info

--envvar, --no-envvar

Show environment variables


Currently, the command line interface (CLI) of Gammapy is implemented using click to define sub-commands, arguments and options, as well as calling the right function that implements a given sub-command.

We have chosen to implement all functionality via a single command line tool called gammapy, with each task as a subcommand. The gammapy command line tool uses the setuptools console_scripts entry point method to automatically create command line tools when Gammapy is installed.

This means that to be able to use the tools you have to install Gammapy. Although, from the source folder you can still execute it without installing via

$ python -m gammapy

which executes gammapy/ as a script as explained here.

Another way to install the gammapy command line tool once, but to have it point at the Gammapy git source folder while you’re hacking on Gammapy is to use

$ python -m pip install --editable .

Either gammapy or python -m gammapy call the gammapy.scripts.main.cli function, which is a click.Group object. If you want you can also import and execute it yourself:

>>> from gammapy.scripts.main import cli
>>> type(cli)
>>> cli()
>>> cli(['--version'])
>>> cli(['image', 'bin', '--help'])

This is what we do to test the CLI, we import gammapy.scripts.main.cli and run it via gammapy.utils.testing.run_cli and check the return code and sometimes console output or generated files. Note that this means that all tests run in a single Python process, we don’t “shell out” and create a subprocess that calls gammapy from a sub-shell.

This is also how the auto-generated CLI documentation in the Reference section above was generated: we use the sphinx-click Sphinx extension that imports and inspects gammapy.scripts.main.cli to find out about all the available sub-commands and help text / arguments / options. The click.Group object exposes all information as attributes and methods. To just give one example:

>>> from gammapy.scripts.main import cli
>>> cli.commands
{'check': <click.core.Group at 0x112107048>,
 'image': <click.core.Group at 0x112102e10>,
 'info': <click.core.Command at 0x1120bb278>}

If you’re new to Python command line tools or Click, probably setuptools entry points and click groups seem very complex. And they are, compared to the rest of Gammapy which is just def and class statements to make functions and classes, i.e. “normal” Python code. Just know that to use and even work on the Gammapy CLI you don’t have to understand how it works under the hood, but if you want to, it’s actually not that complex.

A good path to learn is to start by reading gammapy/ and gammapy/scripts/ and then to look at an example of how a sub-command in Gammapy is implemented and tested, e.g. gammapy/scripts/ and gammapy/scripts/tests/

Note how sub-commands are click.Command objects:

>>> from gammapy.scripts.image_bin import cli_image_bin
>>> type(cli_image_bin)

that are independent and how the main cli is created via cli.add_command calls in gammapy/scripts/

Now you have a basic understanding how things work and should be able to work on Gammapy CLI (e.g. add more functionality to the CLI interface). If you’re curious to learn how it works in detail, we suggest you read the click docs and play with the Gammapy cli object in IPython like we did above when looking at cli.commands, or read and play with the standalone example in the Write your own CLI section below.


The current click-based Gammapy CLI is pretty nice, it is very simple to add commands and also documenting and testing them is pretty nice.

However, the current implementation has some issues and limitations. We describe them in this section, and then in the next one discuss more generally the plan and options for the Gammapy high-level (CLI or non-CLI) interface.

  • There is no support for in-memory tool chain analysis pipelines. I mean something lik e.g. gammapy bin followed by gammapy fit without writing intermediate files and starting the two commands as separate processes.
  • There is no support for configuring or writing provenance information for commands or command pipelines (i.e. store which commands were executed with which arguments in input config or “workflow” files as well as output result files).
  • More generally, we have to see if the separation of Gammapy as a library of “normal” Python functions and classes that can’t be driven by config files or the command line, and then separate functions that represent the CLI is what we want. It means that we have two different ways to use Gammapy in very different ways, and there is duplication and not a nice transition and re-use between the two ways.

More technical issues that can certainly be fixed if we want to stick with the current click-based CLI:

  • gammapy always imports all code from all sub-commands, which drags in large fraction of astropy and gammapy whether it is used or not. This means that gammapy --help takes a few seconds, whereas other commands like git --help just take a very small fraction of a second. This can be improved either by optimising import times throughout Astropy and Gammapy in general, or by lazy-loading the subcommands or by delaying imports into the callbacks from the subcommands.
  • The CLI documentation isn’t nice yet (see the Reference section above). The sphinx-click package that we use isn’t very well-developed or configurable. However, it’s a single Python file that we could just copy into Gammapy and modify and extend to generate documentation in exactly the way we like. E.g. we probably would want to have help text including examples and links to other parts for the Gammapy API and CLI docs that appear nicely on the console as well on in the HTML docs.


There is no concrete plan yet concerning the high-level user interface for Gammapy. Feedback from users and developers on the mailing list is highly welcome! What do you want?

Some options we are considering to build the high-level end-user interface for Gammapy:

  1. Collection of command line tools. Examples: FTOOLs, Fermi ScienceTools, ctools
  2. Config-file driven analysis. Examples: FermiPy, or the H.E.S.S.-internal HAP
  3. No special config- or CLI interface, just normal Python functions and classes. Examples: Sherpa and most Python package like e.g. Astropy or scikit-learn.
  4. Something more fancy that supports tool configuration and running from Python, config files or a CLI using a single implementation. Examples: ctapipe, fact-tools, python-fire

Options 1 and 2 are nice and simple, and they are user-friendly interfaces, and they would allow us to have a stable high-level interface while being able to continue to improve the Gammapy package without breaking user scripts over the coming years.

Their drawback is that they create a second way to use Gammapy, with some users learning and using the more flexible and powerful Python package, and some the simpler to use, but less flexible high-level interface. And note that eventually this second interface will grow into a config file with 100 options (that’s what we have in HAP) or into 10s of CLI tools with in total again 100s of options (see the ctools). However, the Fermi Science tools CLI and Fermipy config interface are examples where the interface size remains at a reasonable level (for users to learn and for developers to implement and maintain) while still exposing everything that most users need.

Options 3 and 4 are similar, in either case the analysis functionality that is available in Gammapy would be written once. Option 3 is what we have now in the Gammapy Python package. Changing to option 4 would mean adding some boilerplate code everywhere (e.g. Python decorators or sub-classing from “tool” base classes like what ctapipe is developing) or relying on the dynamic and inspection features of the Python language offers (see e.g. python-fire), to make it possible to configure and drive analyses not just via Python code, but via some configuration coming either from configuration files (YAML or XML) or command line options.

So what should we do for Gammapy?

I would suggest we continue with the prototyping of the CLI interface as well as of a config-file based interface (gammapy.scripts.SpectrumAnalysisIACT is a starting point). Pull requests that improve and extend what we have are welcome any time!

In parallel, we continue to learn and evaluate solutions others have developed. In python-cli-examples I have started an exploration and evaluation of Python CLI packages, namely click, but also cliff and traitlets, and I might take a closer look also at cement and python-fire there. We should also look in detail at what other projects like LSST, JWST, Fermi, ctapipe, and others do concerning configuration and interface of their science tools. Later in 2018 we will need a comprehensive proposal for code organisation and high-level interface for Gammapy. Suggestions or even contributions are welcome any time!

Write your own CLI

This section explains how to write your own command line interface (CLI).

We will focus on the command line aspect, and use a very simple example where we just call gammapy.stats.significance.

From the interactive Python or IPython prompt or from a Jupyter notebook you just import the functionality you need and call it, like this:

>>> from gammapy.stats import significance
>>> significance(n_observed=10, mu_background=4.2, method='lima')

If you imagine that the actual computation involves many lines of code (and not just a one-line function call), and that you need to do this computation frequently, you will probably write a Python script that looks something like this:

# Compute significance for a Poisson count observation
from gammapy.stats import significance

n_observed = 10
mu_background = 4.2
method = 'lima'

s = significance(n_observed, mu_background, method)

We have introduced variables that hold the parameters for the analysis and put them before the computation. Let’s say this script is in a file called, then to use it you put the parameters you like and then execute it via:

$ python

If you want, you can also put the line #!/usr/bin/env python at the top of the script, make it executable via chmod +x and then you’ll be able to execute it via ./ if you prefer to execute it like this. This works on Linux and Mac OS, but not on Windows. It is also possible to omit the .py extension from the filename, i.e. to simply call the file significance. Either way has some advantages and disadvantages, it’s a matter of taste. Omitting the .py is nice because users calling the tool usually don’t care that it’s a Python script, and it’s shorter. But omitting the .py also means that some advanced users that open up the file in an editor have a harder time (because the editor might not recognise it as a Python file and syntax highlight appropriately), or more importantly that importing functions of classes from that script from other Python files or Jupyter notebooks is not easily possible, leading some people to rename it or copy & paste from it. We’re explaining these details, because if you work with colleagues and share scripts, you’ll encounter the #!/usr/bin/env python and scripts with and without .py and will need to know how to work with them.

Writing and using such scripts is perfectly fine and a common way to run science analyses. However, if you use it very frequently it might become annoying to have to open up and edit the file every time to use it. In that case, you can change your script into a command line interface that allows you to set analysis parameters without having to edit the file, like this:

$ python --help

  Compute significance for a Poisson count observation.

  The significance is the tail probability to observe N_OBSERVED counts or
  more, given a known background level MU_BACKGROUND.

  --method [lima|simple]  Significance computation method
  --help                  Show this message and exit.

$ python 10 4.2

$ python 10 4.2 --method simple

In Python, there are several ways to do command line argument parsing and to create command line interfaces. Of course you’re free to do whatever you like, but if you’re not sure what to use to build your own CLIs, we suggest you give click a try. Here is how you’d rewrite your as a click CLI:

import click
from gammapy.stats import significance

# You can call the callback function for the click command anything you like.
# `cli` is just a commonly used generic term for "command line interface".
@click.option('--method', type=click.Choice(['lima', 'simple']),
              default='lima', help='Significance computation method')
def cli(n_observed, mu_background, method):
    """Compute significance for a Poisson count observation.

    The significance is the tail probability to observe N_OBSERVED counts
    or more, given a known background level MU_BACKGROUND."""
    s = significance(n_observed, mu_background, method)

if __name__ == '__main__':

As mentioned above in the Implementation, we use click in Gammapy itself. We also use click frequently for our own projects if we choose to add a CLI (no matter if Gammapy is used or not). Putting the CLI in a file called makes it easy to go back to a project after a while and to remember or quickly figure out again how it works (as opposed to just having a bunch of Python scripts or Jupyter notebooks where it’s harder to remember where to edit parameters and which ones to run in which order). One example is the gamma-cat

If you find that you don’t like click, another popular alternative to create CLIs is argparse from the Python standard library. To learn argparse, either read the official documentation, or the PYMOTW argparse tutorial. For basic use cases argparse is similar to click, the main difference being that click uses decorators (@command, @argument, @option) attached to a callback function to execute, whereas argparse uses classes and method calls to create a parser object, and then you have to call parse_args yourself and also pass the args to the code or function to execute yourself. So for basic use cases, but also for more advanced use cases where you define a CLI with sub-commands, argparse can be used just as well, it’s just a little harder to learn and use than click (of course that’s a matter of opinion). Another advantage of choosing Click is that once you’ve learned it, you’ll be able to quickly read and understand, or even contribute to the Gammapy CLI.


Besides the CLI interface, the gammapy.scripts package currently contains a bunch of things that will probably all be removed or rewritten and integrated in other sub-packages of Gammapy, leaving scripts just as the high-level command line script interface for Gammapy.

gammapy.scripts Package

Gammapy command line interface (scripts).


create_bg_observation_list(indir, outdir, …) Make total observation list and filter the observations.
group_observations(outdir, overwrite, test) Group list of observations runs according to observation properties.
make_bg_cube_models(indir, outdir[, …]) Create background cube models from the complete dataset of an experiment.
stack_observations(indir, outdir, overwrite) Stack events for each observation group (bin) and make background model.


BgRateTable(energy_lo, energy_hi, data) Background rate table.
CTAIrf([aeff, edisp, psf, bkg, ref_sensi]) CTA instrument response function container.
CTAObservationSimulation Simulate observation for one IRF and target.
CTAPerf([aeff, edisp, psf, bkg, sens, rmf]) CTA instrument response function container.
ObservationParameters([alpha, livetime, …]) Container for observation parameters.
Psf68Table(energy_lo, energy_hi, data) Background rate table.
SensitivityEstimator(irf, livetime[, slope, …]) Estimate differential sensitivity for the ON/OFF method, ie 1D analysis
SensitivityTable(energy_lo, energy_hi, data) Sensitivity table.
SingleObsImageMaker(obs, empty_image, …[, …]) Compute images for one observation.
SpectrumAnalysisIACT(observations, config) High-level analysis class to perform a full 1D IACT spectral analysis.
StackedObsImageMaker([empty_image, …]) Compute stacked images for many observations.
Target([name, model]) Observation target information.