Hi Edward,
The grid_search seems to be working well for the S2 parameters but the ts
fails, returning zero to all spins after the first one. It is strange
because after grid_search finishes the first spin you no longer see the
values of the ts changing. In the end S2 and S2s seem fine but ts returns
zeros for all other spins that are not the first of the sequence.
Regards,
Tiago
-----Original Message-----
From: edward.dauvergne@xxxxxxxxx [mailto:edward.dauvergne@xxxxxxxxx] On
Behalf Of Edward d'Auvergne
Sent: terça-feira, 12 de Abril de 2011 16:27
To: Tiago Pais
Cc: relax-devel@xxxxxxx
Subject: Re: [bug #18019] Gridsearch appears not to be working when
upper/lower bounds are set
Hi,
A grid search can find close to the minimum. If it is fine enough, it
should find it reasonably well. The only problem is if the space is
highly convoluted, and then a course grid search can be way off. This
happens from model m5 a lot, unfortunately. Have a look at my
"Optimisation of NMR dynamic models I" paper for a description of this
problem (http://www.nmr-relax.com/refs.html).
Another problem in this case the assumption of a fast motion at S2f =
0.111 and tf = 0, and then a single motion on top of that, but that is
not something I have tackled yet. Do you have a reference for doing
this?
Regards,
Edward
On 12 April 2011 17:15, Tiago Pais <tpais@xxxxxxxxxxx> wrote:
Yes,
I was just playing with those values also and I have just seen that even a
small increase is already giving different results. As for the zooming
grid
search I don't really know how to implement it and it would probably take
more time than I can spare at the moment, unfortunately.
One more question: will grid_search find the real minimum?
Thanks for all the effort.
Cheers
Tiago P
-----Original Message-----
From: edward.dauvergne@xxxxxxxxx [mailto:edward.dauvergne@xxxxxxxxx] On
Behalf Of Edward d'Auvergne
Sent: terça-feira, 12 de Abril de 2011 16:02
To: Tiago Pais
Cc: relax-devel@xxxxxxx
Subject: Re: [bug #18019] Gridsearch appears not to be working when
upper/lower bounds are set
Hi,
You might just need to increase the number of grid points to maybe [1,
2001, 2001]. If you don't implement a zooming grid search, you may
need to increase the points even more! And the limits for ts should
probably be increased to the nanosecond timescale - trust me, you will
see motions at that scale! I'm running the following command now and
it's looking quite good:
relax> grid_search(lower=[0.111, 0, 0.0], upper=[0.111, 1.0, 2e-9],
inc=[1, 2001, 2001], constraints=True, verbosity=1)
It does take a while without the zooming though. As for those skipped
points, they are valid. The S2 value is less than S2f hence, if used,
would mean that S2s is >1. If the constraints are violated, the grid
points are skipped to speed up the computation time.
Regards,
Edward
On 12 April 2011 16:33, Tiago Pais <tpais@xxxxxxxxxxx> wrote:
It seems to work fine for the first spin but it seems to go crazy on the
following spins (although it seems strange that the S2 value in the array
is
always the same):
Fitting to spin '#snPHSinDiffFrame2CHnoHs_mol1:14&:L@119&@CD1'
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Grid search
~~~~~~~~~~~
Searching through 256 grid nodes.
k: 0 xk: [ 0.111, 0, 0] fk:
16513.3641708
k: 1 xk: [ 0.111, 0.066667, 0] fk:
10390.3333598
k: 2 xk: [ 0.111, 0, 13.333] fk:
7092.56902759
k: 3 xk: [ 0.111, 0.066667, 13.333] fk: 4780.221211
k: 5 xk: [ 0.111, 0.066667, 26.667] fk:
2820.12641286
k: 7 xk: [ 0.111, 0.066667, 40] fk:
2039.04303462
k: 9 xk: [ 0.111, 0.066667, 53.333] fk: 1700.9699681
k: 11 xk: [ 0.111, 0.066667, 66.667] fk:
1546.15216207
k: 13 xk: [ 0.111, 0.066667, 80] fk:
1476.32033148
k: 15 xk: [ 0.111, 0.066667, 93.333] fk:
1454.68522627
Fitting to spin '#snPHSinDiffFrame2CHnoHs_mol1:37&:L@506&@CD1'
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Grid search
~~~~~~~~~~~
Searching through 256 grid nodes.
k: 0 xk: [ 0.111, 0, 0] fk:
5632.02607401
k: 1 xk: [ 0.111, 0.066667, 0] fk:
4534.98672847
And if I increase the verbosity you can see that it just goes through a
lot
of nodes that violate the constraint and the grid point is skipped:
k: 2 xk: [ 0.111, 0.066667, 0]
Constraint violated, skipping grid point.
ci: array([ 1.11000000e-01, 8.89000000e-01, 1.00000000e+00,
0.00000000e+00, -8.89000000e-01, 1.33333333e+14,
-1.33333333e+14])
k: 2 xk: [ 0.111, 0.066667, 0]
Constraint violated, skipping grid point.
ci: array([ 1.11000000e-01, 8.89000000e-01, 0.00000000e+00,
1.00000000e+00, 1.11000000e-01, 1.46666667e+14,
-1.46666667e+14])
Cheers
TP
-----Original Message-----
From: edward.dauvergne@xxxxxxxxx [mailto:edward.dauvergne@xxxxxxxxx] On
Behalf Of Edward d'Auvergne
Sent: terça-feira, 12 de Abril de 2011 15:06
To: Tiago Pais
Cc: relax-devel@xxxxxxx
Subject: Re: [bug #18019] Gridsearch appears not to be working when
upper/lower bounds are set
Hi,
The relax version is the most recent, and the --info print out
indicates that the minfx used is correct (assuming the new version of
relax is at /usr/local/relax-1.3). This is quite peculiar :S If you
use the grid search command:
relax> grid_search(lower=[0.111, 0, 0.0], upper=[0.111, 1.0, 200e-12],
inc=[1, 16, 16], constraints=True, verbosity=1)
What do you see? Try copying and pasting this line into the script.
Cheers,
Edward
On 12 April 2011 15:39, Tiago Pais <tpais@xxxxxxxxxxx> wrote:
Sorry Edward, still not working...
The checked out revision of the repository relax version is number 12819
The errors appear to be exactly the same. What kind of values are you
getting for S2 and tauS? Here, I am getting 0 for all tauS and 0.06667
for
all S2.
$relax --info plots the following:
relax repository checkout
Molecular dynamics by NMR data analysis
Copyright (C) 2001-2006 Edward d'Auvergne
Copyright (C) 2006-2011 the relax development
team
This is free software which you are welcome to modify and redistribute
under
the conditions of the
GNU General Public License (GPL). This program, including all modules,
is
licensed under the GPL
and comes with absolutely no warranty. For details type 'GPL' within
the
relax prompt.
Assistance in using the relax prompt and scripting interface can be
accessed
by typing 'help' within
the prompt.
Hardware information:
Machine: i686
Processor:
System information:
System: Linux
Release: 2.6.31-23-generic
Version: #74-Ubuntu SMP Mon Feb 28 21:32:57 UTC 2011
GNU/Linux version: Ubuntu 9.10 karmic
Distribution: Ubuntu 9.10 karmic
Full platform string:
Linux-2.6.31-23-generic-i686-with-Ubuntu-9.10-karmic
Software information:
Architecture: 32bit ELF
Python version: 2.6.4
Python branch: tags/r264
Python build: r264:75706, Dec 7 2009 18:45:15
Python compiler: GCC 4.4.1
Python implementation: CPython
Python revision: 75706
Numpy version: 1.3.0
Libc version: glibc 2.4
Python packages (most are optional):
Package Installed Version Path
minfx True Unknown
/usr/local/relax-1.3/minfx
bmrblib False
numpy True 1.3.0
/usr/lib/python2.6/dist-packages/numpy
ScientificPython True 2.8
/usr/lib/python2.6/dist-packages/Scientific
wxPython False
mpi4py False
epydoc False
optparse True 1.5.3
/usr/lib/python2.6/optparse.pyc
readline True
/usr/lib/python2.6/lib-dynload/readline.so
profile True
/usr/lib/python2.6/profile.pyc
bz2 True
/usr/lib/python2.6/lib-dynload/bz2.so
gzip True
/usr/lib/python2.6/gzip.pyc
os.devnull True
/usr/lib/python2.6/os.pyc
Compiled relax C modules:
Relaxation curve fitting: True
Sorry for all the trouble.
Cheers
Tiago P
-----Original Message-----
From: edward.dauvergne@xxxxxxxxx [mailto:edward.dauvergne@xxxxxxxxx] On
Behalf Of Edward d'Auvergne
Sent: terça-feira, 12 de Abril de 2011 13:02
To: Tiago Pais
Cc: relax-devel@xxxxxxx
Subject: Re: [bug #18019] Gridsearch appears not to be working when
upper/lower bounds are set
Hi,
Where is the new one installed? Maybe relax is not using it. Are the
errors 100% the same? It's working for me perfectly with the
truncated data. I removed residue 37, and the grid search gives:
-----
relax> grid_search(lower=[0.111, 0, 0.0], upper=[0.111, 1.0,
2.0000000000000001e-10], inc=[1, 16, 16], constraints=True,
verbosity=1)
Over-fit spin deselection.
RelaxWarning: The spin '#snPHSinDiffFrame2CHnoHs_mol1:37&:L@506&@CD1'
has been deselected because of missing relaxation data.
Only the model-free parameters for single spins will be used.
Fitting to spin '#snPHSinDiffFrame2CHnoHs_mol1:14&:L@119&@CD1'
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Grid search
~~~~~~~~~~~
Searching through 256 grid nodes.
k: 0 xk: [ 0.111, 0, 0] fk:
16513.3641708
k: 1 xk: [ 0.111, 0.066667, 0] fk:
10390.3333598
k: 2 xk: [ 0.111, 0, 13.333] fk:
7092.56902759
k: 3 xk: [ 0.111, 0.066667, 13.333] fk: 4780.221211
k: 5 xk: [ 0.111, 0.066667, 26.667] fk:
2820.12641286
k: 7 xk: [ 0.111, 0.066667, 40] fk:
2039.04303462
k: 9 xk: [ 0.111, 0.066667, 53.333] fk:
1700.9699681
k: 11 xk: [ 0.111, 0.066667, 66.667] fk:
1546.15216207
k: 13 xk: [ 0.111, 0.066667, 80] fk:
1476.32033148
k: 15 xk: [ 0.111, 0.066667, 93.333] fk:
1454.68522627
-----
Maybe you need to newest version of relax as well. Try:
$ svn co http://svn.gna.org/svn/relax/1.3 relax-1.3
$ cd relax-1.3
$ svn co http://svn.gna.org/svn/minfx/trunk/minfx
$ cd ..
$ scons
and then see if things run?
Cheers,
Edward
On 12 April 2011 13:49, Tiago Pais <tpais@xxxxxxxxxxx> wrote:
Hi,
I have deleted the previous minfx and installed the new one but It's
still
not working. I also tried to increase the interval of S2f and put some
increments but the same error occurs.
Did you manage to do it with the truncated data?
TiagoP
-----Original Message-----
From: edward.dauvergne@xxxxxxxxx [mailto:edward.dauvergne@xxxxxxxxx] On
Behalf Of Edward d'Auvergne
Sent: terça-feira, 12 de Abril de 2011 10:27
To: Tiago Pais
Cc: relax-devel@xxxxxxx
Subject: Re: [bug #18019] Gridsearch appears not to be working when
upper/lower bounds are set
Hi,
I found the problem to be in minfx. Strangely, this used to work
before as I had performed grid searches with an increment of 1 in one
of the dimensions. The problem is that when there is an increment of
one, you need to pick a value between the lower and upper bound
values. Minfx is now taking the average value.
To use this, you can delete the minfx directory in the 1.3.8
installation, and get the new version of minfx by typing:
$ svn co http://svn.gna.org/svn/minfx/trunk/minfx
If you don't have access to svn, you might have to wait until I
release a new version of relax. However due to holidays, I will
unfortunately not be able to do this before mid May.
Regards,
Edward
On 12 April 2011 10:38, Tiago Pais <tpais@xxxxxxxxxxx> wrote:
Hi Edward,
Yes I am sorry about the private label. I set it initially to private
because I was going to include the structure of the protein attached,
but
then I realized that I didn't need to send the whole thing. Eventually
I
forgot to set the label back to public. So, yes, you can use the info
as
you
wish to try and solve the issue - including the test suite.
Hope to hear from you soon.
Cheers
TP
-----Original Message-----
From: Edward d'Auvergne [mailto:edward.dauvergne@xxxxxxxxx]
Sent: terça-feira, 12 de Abril de 2011 7:51
To: anonymous
Cc: Tiago Pais; relax-devel@xxxxxxx
Subject: Re: [bug #18019] Gridsearch appears not to be working when
upper/lower bounds are set
Hi,
Is there any reason this bug report has been made private? The data
is so truncated, there's nothing much anyone can do with it. The
privacy tag is usually reserved for security issues whereby it needs
to be kept private until a fix is found and released. Then the bug
can be made public. I'm looking into this now, but I'm not exactly
sure what is happening. Is it ok if I include this data, or part of
it, into the relax test suite? This will significantly help in the
debugging and will make sure this bug never returns!
Cheers,
Edward
On 11 April 2011 13:42, anonymous <NO-REPLY.INVALID-ADDRESS@xxxxxxx>
wrote:
URL:
<http://gna.org/bugs/?18019>
Summary: Gridsearch appears not to be working when
upper/lower bounds are set
Project: relax
Submitted by: None
Submitted on: Mon 11 Apr 2011 11:42:49 AM UTC
Category: relax's source code
Severity: 3 - Normal
Priority: 5 - Normal
Status: None
Privacy: Private
Assigned to: None
Originator Name: Tiago Pais
Originator Email: tpais@xxxxxxxxxxx
Open/Closed: Open
Discussion Lock: Any
Release: 1.3.8
Operating System: GNU/Linux
_______________________________________________________
Details:
My system is Ubuntu-9.04:
When I perform a gridsearch, exclusively for m5, with the lower and
upper
bounds set to: lower=[0.111, 0, -0.0], upper=[0.111, 1.00, 200] I
keep
seeing
the following message during the calculations:
"Min f: 3474.82
k: 3860 xk: [ 0.111, 0.066667, -0] fk: nan
Increment: array([ 5, 16, 16])
Params: array([ 0.111, 1. , Inf])
Min params: array([ 0.111 , 0.06666667, -0. ])
f: nan
Min f: 3474.82
k: 3861 xk: [ 0.111, 0.066667, -0] fk: nan
Increment: array([ 6, 16, 16])
Params: array([ 0.111, 1. , Inf])
Min params: array([ 0.111 , 0.06666667, -0. ])
f: nan
"
And at the end the tauS is zero for all spins and the S2s is always
close
to
1.
Attached follows the script that I am using and a truncated sample of
my
data
set (I hope it is not too truncated!) as well as the results that I
am
getting.
If you need more info let me know.
Thanks
Regards
Tiago P
_______________________________________________________
File Attachments:
-------------------------------------------------------
Date: Mon 11 Apr 2011 11:42:49 AM UTC Name: ErrorReportRelax.tar.gz
Size:
8kB By: None
<http://gna.org/bugs/download.php?file_id=12833>
_______________________________________________________
Reply to this item at:
<http://gna.org/bugs/?18019>
_______________________________________________
Message sent via/by Gna!
http://gna.org/
_______________________________________________
relax (http://nmr-relax.com)
This is the relax-devel mailing list
relax-devel@xxxxxxx
To unsubscribe from this list, get a password
reminder, or change your subscription options,
visit the list information page at
https://mail.gna.org/listinfo/relax-devel
__________ Information from ESET NOD32 Antivirus, version of virus
signature
database 6034 (20110411) __________
The message was checked by ESET NOD32 Antivirus.
http://www.eset.com
__________ Information from ESET NOD32 Antivirus, version of virus
signature
database 6034 (20110411) __________
The message was checked by ESET NOD32 Antivirus.
http://www.eset.com
__________ Information from ESET NOD32 Antivirus, version of virus
signature
database 6034 (20110411) __________
The message was checked by ESET NOD32 Antivirus.
http://www.eset.com
__________ Information from ESET NOD32 Antivirus, version of virus
signature
database 6035 (20110412) __________
The message was checked by ESET NOD32 Antivirus.
http://www.eset.com
__________ Information from ESET NOD32 Antivirus, version of virus
signature
database 6035 (20110412) __________
The message was checked by ESET NOD32 Antivirus.
http://www.eset.com
__________ Information from ESET NOD32 Antivirus, version of virus
signature
database 6035 (20110412) __________
The message was checked by ESET NOD32 Antivirus.
http://www.eset.com
__________ Information from ESET NOD32 Antivirus, version of virus
signature
database 6035 (20110412) __________
The message was checked by ESET NOD32 Antivirus.
http://www.eset.com
__________ Information from ESET NOD32 Antivirus, version of virus
signature
database 6035 (20110412) __________
The message was checked by ESET NOD32 Antivirus.
http://www.eset.com
__________ Information from ESET NOD32 Antivirus, version of virus
signature
database 6036 (20110412) __________
The message was checked by ESET NOD32 Antivirus.
http://www.eset.com
__________ Information from ESET NOD32 Antivirus, version of virus
signature
database 6036 (20110412) __________
The message was checked by ESET NOD32 Antivirus.
http://www.eset.com
__________ Information from ESET NOD32 Antivirus, version of virus signature
database 6036 (20110412) __________
The message was checked by ESET NOD32 Antivirus.
http://www.eset.com
__________ Information from ESET NOD32 Antivirus, version of virus signature
database 6037 (20110412) __________
The message was checked by ESET NOD32 Antivirus.
http://www.eset.com
Re: [bug #18019] Gridsearch appears not to be working when upper/lower bounds are set