This repository is designed for easily running Amiga Halo Finder on the FIRE simulations. It uses a significant amount of code written by Sasha Muratov.
- Clone the repository with
git clone git@bitbucket.org:zhafen/ahf_wrapper.git. - Move into the AHF subdir (e.g.
ahf-v1.0-069), and make AHF (i.e.make veryclean; make AHF MergerTree MergerTrace). - Install my general analysis repository galaxy-dive (
pip install galaxy-dive).
In the future, I hope to make this installation process more automatic.
To run AHF, simply consider making the following edits to ahf_pipeline.sh, and then submit it as a job.
- Go through each of the slurm job submission options, and make sure that they are relevant. E.g. does the output go where you want? Do you have the queue specified? Are you mailing the right user?
- Change snap_dir to the location of the simulation data.
- Change out_dir to the location you want the halo data output too. Do not send the output to the shared FIRE folder. AHF produces several hundred GB of halo data, in addition to a full copy of the simulation output converted to binary, which can and should be deleted later on.
- Specify where the simulation metafiles are located. Specifically, the two files requires are: snapshot_times.txt and -usedvalues. Make sure there is only one file that ends in "-usedvalues" in the directory. If you're missing a "-usedvalues" file, you can create one from the config file using the helper script convert_to_usedvalues.py.
- Choose the snapshot range you want this use.
- Choose how many processors you want to use.
- Choose if you want to convert snapshots to binary (
convert_snapshots; AHF reads in binary files, so this step is typically essential). - Choose if you want to run AHF itself (
find_halos; also essential). - Choose if you want to correlate halos between files (
find_merger_tree; usually important). - Choose if you want to trace particular halos through the entire simulation (
find_merger_trace; usually important) - Choose if you want to get extra information about each halo not usually included by AHF (
get_ahf_halos_adds; this is optional, and includes information like the stellar half-mass radius, the halo concentration, mass contained within the stellar half-mass radius, and more) (requires Colossus installed). - Choose if you want to smooth halos (
smooth_halos; optional, but useful).
If the pipeline fails, the most common issue is a corrupted snapshot, so check the raw data. Many snapshots directories often contain hidden files (e.g. .snapshot_014.1.hdf5.aKDJLd), which can often also cause problems. The pipeline can also run out of memory if you have too many processes running at once, try reducing n_proc and restarting.
Another common problem are unbalanced snapshot sizes. For example, the original snapshot output by the simulation can sometimes produce sub-snapshots >15GB in size. This will break the pipeline. To get around this, I've written a function that takes the original hdf5 files, loads the snapshot, then saves the data in snapshots with an even file size. This function is part of galaxy_diver, in particular it is galaxy_diver.utils.hdf5_wrapper.copy_snapshot().
Contact Zach Hafen (zachary.h.hafen@gmail.com) with any questions or problems.