This project tries to handle all the steps needed for a typical AGN-analysis (or any point-source really):
- Select runs observing the source and matching some quality criteria
- Apply models on dl1 files / Produce dl2
- Calculate IRFs
- Produce dl3
- Run the gammapy data reduction to dl4
- Perform spectral fit, calculate flux points, calculate light-curve, ...
- Perform multiple gammapy analyses with the same dl3
What it does NOT handle:
- Low-level calibrations. We start at DL1
Conceptually there are three main steps (You could define more, but these are still somewhat split in the workflow):
- Select runs via the datacheck files
- Link runs and mc to the
build
directory. This makes rules easier to reason about. These are only links, not copies (!) and show up as such usingls -l
. - Run the analysis using
lstchain
andgammapy
The different workflow stages are separated into different files where I saw a clear
distinction. This is to keep each file relatively small and reduce the number of variables one
has to reason about at once. The different stages are not directly usable on their own
as they assume a certain directory structure and file naming conventions, but it should be
straightforward to adapt this to ones liking if one would want to build their own analysis.
I try to keep all variables in workflow/definitions.smk
, which should cover most, but
not necessarily all. Also the scripts should all be usable on their own with the small caveat,
that sometimes things get imported from the top level scripts
directory, so
you will need to make sure these things are present as well or copy them there.
You need configuration files. Check our example repository for how to use them, and clone them next to this repo, e.g.
git clone https://github.com/nbiederbeck/lst-analysis-config.git ../lst-analysis-config
You need snakemake and astropy installed. If you do not have that, you can create an enviroment with only these packages like this:
mamba env create -f workflow/envs/snakemake.yml
mamba activate snakemake
For development, you should also install atleast pre-commit.
The enviroment in workflow/envs/data-selection
should contain everything (Don't forget to pre-commit install
afterwards).
This is not a specific development enviroment though.
Cleaning up the envs is definetively a DEV-TODO as it got a bit messy after merging what was originally two projects.
Also you need the source catalogue. Since this requires credentials (standard LST ones), it is not done automatically.
This is needed only once (or whenever you want to use new runs).
You can use this command (replace <username>
and <password>
):
curl --user <username>:<password> \
https://lst1.iac.es/datacheck/lstosa/LST_source_catalog.html \
-o runlist.html
Adapt to your liking
- Data selection:
../lst-analysis-config/data-selection.json
- MCs to use, lstchain env and number of off regions:
../lst-analysis-config/lst_agn.json
(gammapy does not handle energy-dependent cuts automatically, so we need to work around this) - IRFs (lstchain):
../lst-analysis-config/irf_tool_config.json
- gammapy:
analysis.yaml
andmodels.yaml
in subdirectories../lst-analysis-config/analysis_*
. These all use the same dl3, but produce their own dl4
If you want to use configs from a different place, instead call make
using make CONFIG_DIR=/path/to/configs
.
Basically only run make
.
If you run it without a specific target, everything should be resolved.
In the event, that you give a target such as make build/plots/analysis-one/flux_points.pdf
, the linking part might not run again.
This is clearly suboptimal, but for now you will need to run make link
first IF YOU CHANGED SOMETHING in the data selection.
Since in this case, one probably wants to rerun everything anyway, this was not super high on our priority.
It is related to nbiederbeck#26
If you have run snakemake on the cluster, you can create the plots and tex files locally (using your own matplotlibrc for example).
We separate the calculation of metrics and the plotting to make sure you can finetune plots later on without needing to
run the expensive steps on the local machine. The tables for that are saved as either fits.gz
or h5
.
For the data-selection plots, you need to download build/dl1-datacheck-masked.h5
, e.g.:
mkdir -p build
scp <host>:<path-to>/lst-data-selection/build/dl1-datachecks-masked.h5 build/
Afterwards:
make -f local.mk
DEV-TODO: Do the same for the other plots (nbiederbeck#29)
If you do some cp **/*.fits.gz
shenanigans, beware that the dl3 files are saved with
the extension fits.gz
as well.