``` \ / / |-\ ----- \ / | | \ | \ / / | | |-- \ / | | / | \/ / |-/ ----- *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-* ``` This is VIDE, the Void IDentification and Examination toolkit. For more information, see http://www.cosmicvoids.net Please cite arXiv:1406.1191 and arXiv:0712.0349 if you use this software, using the following suggested sentence: "This work uses voids identified with VIDE\footnote{\url{http:www.cosmicvoids.net}} (Sutter et al. 2014), which implements an enhanced version of ZOBOV (Neyrinck 2008) to construct voids with a watershed algorithm." License/Copyright information ----------------------------- Copyright (C) 2010-2020 Guilhem Lavaux, 2011-2014 P.M. Sutter. This software is put under the GNU Public License. Please see LICENSE for further information. Mainline VIDE contributions from Ben Wandelt, Nico Hamaus, Alice Pisani, Paul Zivick, and Qingqing Mao. This toolkit includes ZOBOV, originally developed by Mark Neyrinck. See `zobov/zobov_readme.txt` for copyright/license information. SDF library provided by Michael S. Warren and John Salmon. HOD fitting code provided by Francisco Navarro. HOD halo population code provided by Jeremy Tinker. RAMSES module provided by Benjamin B. Thompson. Requirements ------------ The package swig needs to be installed and available in the PATH (http://www.swig.org/). It is required by scipy and we have not decided to bundle it with VIDE at the moment. Quick Start Guide ----------------- After installing the package with `python3 setup.py install --user`, you can execute ``` python3 -m void_pipeline your_config_file.py ``` The VIDE tools are all packaged in the `vide` package. Running with simulation ----------------------- !! Temporary measure !! After build the vide package and installing it with `python3 setup.py install` (for example), it is possible to find the script to build the pipeline for simulations in `build/temp.[SOMETHING]/pipeline/prepareInputs.py`. Just copy that script where you want the analysis and execute it directly with the parameter file as an argument. For example: ``` mkdir $HOME/my_vide_test cp build/temp.[SOMETHING]/pipeline/prepareInputs.py $HOME/my_vide_test cp python_tools/void_pipeline/datasets/example_simulation.py $HOME/my_vide_test mkdir $HOME/my_vide_test/examples cp examples/example_simulation_z0.0.dat $HOME/my_vide_test/examples cd $HOME/my_vide_test python3 prepareInputs.py --all --parm example_simulation.py python3 -m void_pipeline example_simulation/sim_ss1.0.py ``` Notes for CONDA --------------- If you use a conda installation, you have to be sure to use all the building tools that are consistent. On linux that means for example installing the conda packages `gcc_linux-64` and `gxx_linux-64`. In addition to that it is recommended to define the environment variable `LIBRARY_PATH=the_path_of_your_conda_environment_with_/lib`. For example if your environment is in '/home/user/conda' you should define ``` export LIBRARY_PATH=/home/user/conda/lib ``` You can then initiate the construction with ``` python3 setup.py build ``` Version Summary ----------------- v1.0 - Initial Release v2.0 - Ported to python3, revisited build system