Hi Bernhard,
I realized this when comparing the functions yesterday. Exactly like what
you said, when a user interrupts the process, what he/she wants at the
moment is most likely to be going back a few residues instead of starting
all over again. I am quite happy to have both interruptible and
uninterruptible functions to choose from. Thanks for the effort.
But maybe it would make the impatient ones happier by presenting both
versions in the Extension menu?
Besides speed, one other advantage I can see for the old function is that by
having a simpler structure, it is way easier for the users to make their own
modifications. May I beg you not to remove it from future versions?
Regards,
Zhijie
--------------------------------------------------
From: "Bernhard C. Lohkamp" <bernh...@chem.gla.ac.uk>
Sent: Thursday, March 31, 2011 1:11 PM
To: <CCP4BB@JISCMAIL.AC.UK>
Subject: Re: [ccp4bb] step refine speed of wincoot - a little drawback of
the new interruptible_fit_protein() function in wincoot
Thank you for the comment. You are correct that there are being backups
made in the newer interruptable fit function (they were't previously in
the uninterruptable fitting). I wasnt thinking about that (*). This may
(especially on windows) slow down things a great deal as we write and
gzip these files as well. To note here, gzip may be slow on windows as
is writing files. In the pre-release versions I disabled the gzip on
windows by default (and in general made the compression of backup files
optional). This may speed up things (in certain circumstances).
I hope this clarifies things a bit...
B
(*) It's debatable if we want backups here or not. If you can and want
to interrupt the fitting you may want to go back one step, which
requires a backup. On the other side this will write a lot of backup
files which is slowing thing down by writing and compressing files, as
well as producing a lot of backup files. Discuss... (or Paul and I just
decide).
Hi Xiaopeng, and those who are using the new interruptible
"fit_protein..." or "stepped_refine..." in wincoot 0.6.1:
I just took a look at wincoot 0.6.1's fitting.py file
(WinCoot\share\coot\python\fitting.py). It seems that in the old
"fit_protein()" and "stepped_refine_protein()" functions, the backup
was programmed to be turned off during the cycles when it steps
through the protein.
However when using the "Stepped refine..." from the "Extensions" menu,
the program starts to spit out tons of backup files. I then checked
the "extension.py". It turns out that these menu items are calling the
new "interruptible_fit_protein()" function instead of the old
uninterruptible ones. Some further experiment showed that this
behavior was dependent on the "backup state" variable of the model
being refined. When the "backup state" is 1 (the default value?),
these new interruptible fitting functions will save a compressed
backup file on each step.
So, in order to save time when using the new interruptible protein
fitting functions, one can type the following command in the python
script window to tweak the "backup state" to 0:
make_backup(imol) #to save a backup
turn_off_backup(imol)
where imol is the number associated with the pdb file being fitted.
Do not forget to turn it back on after the fitting:
turn_on_backup(imol)
To check the current model's "backup state":
backup_state(imol)
Zhijie
--------------------------------------------------
From: "Xiaopeng Hu" <huxp...@mail.sysu.edu.cn>
Sent: Tuesday, March 29, 2011 9:02 AM
To: "Zhijie Li" <zhijie...@utoronto.ca>
Subject: Re: [ccp4bb] step refine speed of wincoot
Dear Zhijie,
Thanks for there suggestions.
I tried to turn off backup (with script command) and it did help
although not much. Is there any way to add this perference to setting
of coot/wincoot?
best
xiaopeng
----- 原始邮件 -----
发件人: "Zhijie Li" <zhijie...@utoronto.ca>
收件人: "Xiaopeng Hu" <huxp...@mail.sysu.edu.cn>
发送时间: 星期二, 2011年 3 月 29日 下午 7:46:05
主题: Re: [ccp4bb] step refine speed of wincoot
Hi,
If you feel the refinement to be too slow, you may turn off the smooth
centering (in preferences) or change the centering steps to a smaller
number
to save unnecessary graphical calculations. To go extreme, you may even
remove the real time display commands in the scripts - also a way to
test if
the difference observed is due to different graphical efficiency.
Reducing
the size of the displayed map also helps.
The other thing you may need to consider is that coot/wincoot will
save a
backup file each time it updates the model, which means on each step
of the
refinement you have a new file generated (take a look at the coot-backup
directory when doing stepped refinement). If your windows disk is
terribly
fragmented then sure you will spend a lot of time on writing these
files.
The other thing is, windows has a different file caching mechanism than
linux, this can also cause a huge difference when a few hundred small
temporary files are queued for writing. My impression is that both the
ext2/3 file system and the way linux handles caching are more
efficient for
this kind of situations. You may try deleting the files in
wincoot\coot-backup periodically and defragmenting that partition.
Making a
virtual disk in the RAM to put your backup directory there could be
something to experiment on too.
Zhijie
--------------------------------------------------
From: "Xiaopeng Hu" <huxp...@mail.sysu.edu.cn>
Sent: Sunday, March 20, 2011 9:22 AM
To: <CCP4BB@JISCMAIL.AC.UK>
Subject: [ccp4bb] step refine speed of wincoot
Dear all,
I found the step refine speed of wincoot is much slower than that of
linux
coot (with the same pc). Is it normal or I need to configure something
with the wincoot?
best,
Xiaopeng Hu
