Revision 4f75faeded2cb284dedbc856a8b2ae56075ea158 authored by Collin Capano on 20 June 2020, 18:27:09 UTC, committed by GitHub on 20 June 2020, 18:27:09 UTC
* use different acl for every chain in epsie * create base burn in class, move common functions to there; rename MCMCBurnInTests EnsembleMCMC, first stab at creating MCMC tests for independent chains * more changes to burn in module * simplify the attributes in the burn in classes * add write method to burn in classes * add write_data method to base_hdf * remove write_burn_in method from mcmc io; use the write method in burn in module instead * make use of new burn in functions in sampler/base_mcmc * have emcee and emcee pt use ensemble burn in tests * add compute_acf function to epsie * start separating ensemble and mcmc io methods * stop saving thin settings to file; just return on the fly * make read/write samples stand alone functions, and update emcee * rename write functions; update emcee * move multi temper read/write functions to stand alone and update emcee_pt * pass kwargs from emcee(_pt) io functions * simplify get_slice method * add function to base_mcmc to calculate the number of samples in a chain * use nsamples_in_chain function to calculate effective number of samples * add read_raw_samples function that can handle differing number of samples from different chains * add forgotten import * use write/read functions from base_multitemper in epsie io * use stand alone functions for computing ensemble acf/acls * separate out ensemble-specific attributes in sampler module; update emcee and emcee_pt * add acl and effective_nsample methods to epsie * simplify writing acls and burn in * fix various bugs and typos * use a single function for writing both acl and raw_acls * add some more logging info to burn in * reduce identical blocks of code in burn in module * fix self -> fp in read_raw_samples * reduce code duplication in base io and simplify read raw samples function * fix missed rename * reduce code redundacy in sampler/base_multitemper * whitespace * fix bugs and typos in burn_in module * fix code climate issues * use map in compute_acl * more code climate fixes * remove unused variable; try to silence pylint * fix issues reading epsie samples * only load samples from burned in chains by default * add act property to mcmc files * fix act logging message * fix effective number of samples calculation in epsie * remap walkers option to chains for reading samples * fix thinning update * fix acceptance ratio and temperature data thinning in epsie * allow for different fields to have differing number of temperatures when loading * don't try to figure out how many samples will be loaded ahead of time * store acts in file instead of acls * write burn in status to file before computing acls * drop write_acts function * fix issue with getting specific chains * fix typo * code climate issues * fix plot_acl
1 parent 926b628
INSTALL
PyCBC installation instructions
===============================
Prerequisites
-------------
* LALSuite
See https://www.lsc-group.phys.uwm.edu/daswg/docs/howto/lal-install.html
for instructions to install through source, or simply install the pre-built
version hosted on pypi using pip.
Installing PyCBC
----------------
pip install lalsuite pycbc
Installing CUDA Python modules
------------------------------
For GPU acceleration through CUDA:
* Nvidia CUDA >= 4.0 (driver and libraries).
* PyCUDA >= 2013.1.1 - http://mathema.tician.de/software/pycuda
See next section for instructions.
* SciKits.cuda >= 0.041 - http://scikits.appspot.com/cuda
See next section for instructions.
* Mako >= 0.7.2 - http://www.makotemplates.org
See next section for instructions.
These packages may not be available via the distribution packaging system,
at least in the required versions. Although they should be generally
installable via pip or easy_install, this method is not always available
on LSC clusters. Therefore, we report general instructions for installing
from source on your ~/.local directory.
PyCUDA:
git clone http://git.tiker.net/trees/pycuda.git
cd pycuda
git submodule init
git submodule update
./configure.py
python setup.py build
python setup.py install --user
If your CUDA installation is in a non-standard location X,
pass --cuda-root=X to configure.py.
SciKits.cuda:
Get the tarball (http://pypi.python.org/pypi/scikits.cuda) and unpack it.
cd scikits.cuda*
python setup.py install --user
Mako:
Get the tarball (http://www.makotemplates.org/download.html) and unpack it.
cd Mako*
python setup.py install --user
To install system-wide rather than to ~/.local, change the 'setup.py install'
commands to
sudo python setup.py install
Computing file changes ...
