Module mcmc_m4p
source code
This is a helper module to make use of mcmc.py and mcmc_big.py. It
allows you to conveniently run a Monte-Carlo simulation of any kind until
it converges (stepper.run_to_bottom) or until it has explored a large
chunk of parameter space (stepper.run_to_ergodic).
It also helps you with logging the process.
When run in parallel, each processor does its thing more-or-less
independently. However, every few steps, they exchange notes on their
current progress. If one finds an improved vertex, it will be passed on
to other processors via MPI.
Tested with mpi4py version 1.1.0, http://mpi4py.scipy.org
|
precompute_logp(lop)
Does a parallel evaluation of logp for all items in lop. |
source code
|
|
|
|
|
|
|
|
|
|
|
Debug = 0
|
|
mpi = MPI.COMM_WORLD
|
Imports:
sys,
MPI,
random,
numpy,
die,
mcmc,
MCH,
TooManyLoops,
warnevery,
logger_template,
test,
step_acceptor,
make_stepper_from_lov