To target the grid search, what will need to be done is to have a grid
search of grid searches. This was discussed in the thread starting at
https://mail.gna.org/public/relax-devel/2007-03/msg00088.html,
Message-id: <45FEAC46.8020307@xxxxxxxxxxxxxxx>). I think that this
idea should be targeted specifically to the model-free minimise()
function. It should occur within the main minimisation loop but only
when grid searches are detected. Additionally it should only occur
when the param_set is 'diff' (diffusion parameters variable,
model-free parameters fixed). No grid search should ever be executed
on the 'all' parameter set (diffusion parameters variable, model-free
parameters variable) because the grid search will be way to big.
Currently a RelaxError is thrown for this situation. The grid search
of the param_set of 'mf' and 'local_tm' (diffusion parameters fixed)
should also not be parallelised as there are multiple spin systems to
send out to nodes and the grid search is relatively quick.
Ideally you would be able to detect how many nodes are in the cluster
through mpi4py and then split up the 'diff' model-free grid into
maximally that many blocks. Then a grid search will need to be
carried out. Each element of the grid search loop would run the true
grid search from the minimisation code on a subset of the full grid
and then store the results (parameter vector + min stats). Once all
the elements have been calculated, the minimised parameter vector is
the result from the element which has the lowest chi-squared value.
These changes will necessarily be more extensive than the current
changes to 'specific_fns/model_free.py'.
Cheers,
Edward
On 3/30/07, Gary S. Thompson <garyt@xxxxxxxxxxxxxxx> wrote:
Do you have any thoughts on the question of how to parallelise grid
searches and monte
carlo runs and where to dig into the code base?
regards
gary
note I have already realise one wrinkle about paralleliseing the
montecarlo runs. We need to have consistent (!) use of random numbers on
the various processors. Therefore am going to create a command in
processor that will create and resturn a block of random numbers (I
guess we also need to think about what distributions we need as well)
regards
gary
--
-------------------------------------------------------------------
Dr Gary Thompson
Astbury Centre for Structural Molecular Biology,
University of Leeds, Astbury Building,
Leeds, LS2 9JT, West-Yorkshire, UK Tel. +44-113-3433024
email: garyt@xxxxxxxxxxxxxxx Fax +44-113-2331407
-------------------------------------------------------------------
Re: how to parallelise monte carlo and grid searches