[Dock-fans] "make test" , "possible failure" and its association with operating system

Scott Brozell sbrozell at rci.rutgers.edu
Wed Jul 8 11:27:59 PDT 2009


Hi,

A clarification is that the comparison program does account for
floating point precision, but that is insufficient.
The problem scenario is slight floating point changes that cause
different sampling and/or scoring so that a different ligand pose is
'selected'; this then causes significant differences that cannot be
filtered at that stage by accounting for floating point precision.
This most frequently occurs with tests that involve contact or grid
scoring.  Somewhat surprisingly to me, using a formatted as opposed
to a binary representation for the intermediate files may not
circumvent this problem.  Thus, my workaround, which is progressing
but slowly, is to write new tests that avoid contact or grid scoring
and then to localize all contact or grid scoring tests so that we
can easily flag these as the source of the problem.

The take home message is that a small number of failures is
rarely an indication of real problems, but if almost
every test fails then something is amiss.
In the former case, it may be useful to determine whether the tests
employ contact or grid scoring.

Scott

On Tue, Jul 07, 2009 at 04:49:49PM -0700, John J. Irwin wrote:
> Let me try to be clearer.  Floating point (real) numbers get rounded off
> differently on different processors and different operating systems. I
> am not an expert in these things, but if you use anything other than an
> intel ix86 running a recent linux it is not unusual to see *slight*
> changes in perhaps the 6th or 7th significant figure of floating point
> numbers.  Our comparison program (as I understand it) is not clever
> enough to figure this out. So I suggest you just look at the output
> files and see whether they look as close as you would expect given
> roundoff error.
> 
> John
> 
> 
> Rajeshwari Iyer wrote:
> > Respected Dr. Irwin and other friends,
> > Thank you.
> > I see you you mentioned "possible failure" might be associated with
> > operating system.
> > (question 1) I am using a pc that has Windows XP prof service pack 3
> > operating  system.
> > Intel core duo processor. Do you guys know if docking can be done in
> > this type of system
> > if not (question 2) Is there any operating system for example Linux
> > that would work with dock 6.2
> > I tried one type of Linux but never could do nothing. That is why
> > may I kindly ask for exact specification of a
> > computer system and the operating system its running... if I have to
> > run dock6.2 in Linux.
> > (question 3) If cygwin works for me, will I be able to change source
> > code etc. in cgwin env. or cygwin is
> > simply a running envoirment. (question 4) For cygwin, I did " make
> > test" ...before that "make"
> > ......do I need to do "make all" or "make dock" or make utils" to do
> > docking? 


More information about the Dock-fans mailing list