Hi, please see my update below

On Friday, January 16, 2015 at 9:44:49 PM UTC, abrukhno wrote:
> On Friday, January 16, 2015 at 7:58:01 PM UTC, Christoph Junghans wrote:
> > 2015-01-16 9:46 GMT-07:00 abrukhno <[email protected]>:
> > > Hi,
> > >
> > > I have been amazed how long (the entire night) it takes to finish an 
> > > IMC+Octave iteration step after the simulation done.
> > >
> > > With a simulation taking under an hour and IMC Octave solver taking 
> > > almost exactly 12 hours, one iteration step for a pretty small system 
> > > (but millions frames) takes just over half a day.
> > >
> > > Same type of IBI iteration step takes just about 3 hours.
> > >
> > > Tried for IMC: <group>all</group> and splitting a set of 10 potentials 
> > > into 3 groups, with virtually same times.
> > >
> > > Is it normal? Might Numpy improve on the speed?
> 
> > Did octave took a long time to run or csg_stat?

- it is certainly csg_stat, as the progress of files show:
 
- Here is how it looks like:
 drwxr-xr-x  44K Nov  8 09:33 step_001/
 drwxr-xr-x  44K Nov  8 22:30 step_002/
 drwxr-xr-x  44K Nov  9 11:41 step_003/
 drwxr-xr-x  44K Nov 10 00:46 step_004/
 drwxr-xr-x  44K Nov 10 13:55 step_005/
 
 and inside a step:
 
 -rw-r--r--  307K Nov 11 02:51 table_CH_CH.xvg
 -rw-r--r--  307K Nov 11 02:51 table_CH_CO.xvg
 -rw-r--r--  4.9M Nov 11 03:44 confout.gro
 -rw-r--r--  1.3M Nov 11 03:44 ener.edr
 -rw-r--r--   92K Nov 11 03:44 md.log
 -rw-r--r--  2.9K Nov 11 15:45 CH-CH.dist.new
 -rw-r--r--  2.9K Nov 11 15:45 CH-CO.dist.new
 
In my recent attempts I have a longer trajectory and up to now I have not seen 
yet the distributions produced by csg_stat, after 18 hours. Below are a few 
latest lines in inverse.log:
===
...
table_extrapolate.pl: extrapolating the left using linear with gradient 0
table_extrapolate.pl: extrapolating the right using constant with gradient 0
begin to calculate inverse monte carlo parameters
# of bonded interactions: 0
# of non-bonded interactions: 10
===
It appears csg_stat for IMC is a lot more time consuming than for IBI. You 
probably know it, but the surprise for me is, it takes so unbearably long, 
making IMC virtually useless. 

> > The octave runtime depends a lot on the step size (=size of the matrices).
> > 3 groups triple the time for solving.
> 
> - wow, I probably stopped that after half a day
>  
> > VOTCA supports numpy and matlab for solving the equations, too, but I
> > don't think it will make a huge difference in runtime.
> 
> - thought so..
> 
> > In the original VOTCA paper, we did a couple of error comparisons for
> > IBI vs. IMC.
> > IMC is usually better for the first couple of iterations, but needs
> > more statistics than IBI.
> 
> - 5x/10x?

- my experience is the more stats you have the better the result with either of 
these (whence my long trajectories).

Thank you for reply, Christoph!
 
> Will try to investigate it further.
> 
> > 
> > Christoph
> > >
> > > Thank you for your clues.
> > >
> > > Andrey
> > >
> > > --
> > > You received this message because you are subscribed to the Google Groups 
> > > "votca" group.
> > > To unsubscribe from this group and stop receiving emails from it, send an 
> > > email to [email protected].
> > > To post to this group, send email to [email protected].
> > > Visit this group at http://groups.google.com/group/votca.
> > > For more options, visit https://groups.google.com/d/optout.
> > 
> > 
> > 
> > -- 
> > Christoph Junghans
> > Web: http://www.compphys.de

-- 
You received this message because you are subscribed to the Google Groups 
"votca" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/votca.
For more options, visit https://groups.google.com/d/optout.

Reply via email to