Dear all,
I agree with Steve that the results shouldn't change. If you explicitly
give the same random seed (--random_seed) then this should give the exact
same results. Or, at least the differences should be very small. I think
the betagal data from the tutorial should reach full Nyquist.
HTH,
Sjors
> I would be really interested to hear Sjors comments on this. The
> 'resolution improvement' to me sounds very worrying. One thing the 'gold
> standard' method does not account for is any 'algorithm bias'. That is, if
> the algorithm is producing some artifact in a consistent way in both the
> even and odd maps it could result in resolution overestimation even when
> the structure is actually getting worse. Speaking as a programmer, if you
> make a speed optimization to an algorithm and the results change in any
> measurable way that means you did _something_ other than just speed things
> up.
>
> To make a practical suggestion, have you computed an FSC between the final
> map computed using the original code and the optimized version?
>
> ----------------------------------------------------------------------------
> Steven Ludtke, Ph.D.
> Professor, Dept. of Biochemistry and Mol. Biol. Those who
> do
> Co-Director National Center For Macromolecular Imaging ARE
> Baylor College of Medicine The
> converse
> [log in to unmask]<mailto:[log in to unmask]> -or-
> [log in to unmask]<mailto:[log in to unmask]>
> also applies
> http://ncmi.bcm.edu/~stevel
>
> On Jan 25, 2016, at 5:14 PM, Sargis Dallakyan
> <[log in to unmask]<mailto:[log in to unmask]>> wrote:
>
> Hi Dimitry,
>
> Great, thanks a lot. It compiles and works fine now. I run a quick
> benchmark job for the 3D refinement step from the Relion 1.3 tutorial,
> using the beta-galactosidase data and pre-calculated results for the
> earlier steps, similar to the one presented at
> http://www.ccpem.ac.uk/user_help/relion_benchmarks.php
>
> Hardware used: Intel Xeon Processor E5-2670 v3 (30M Cache, 2.30 GHz), 24
> cores per node, 256GB RAM per node, FDR infiniband (purchased 2015).
>
> Code used
> # Nodes # MPI procs # total threads Wallclock
> time CPU time
>
> Relion 1.4
> 10 200 400 00:33:14 107:09:56
>
> vlion 618c81e
> <https://github.com/dtegunov/vlion/commit/618c81e2beb27a5c35847c33710ac1d0ea5ad25c>
> 10
> 200
> 400
> 00:26:54
> 85:23:50
>
> I've used tracejob command to get Wallclock and CPU time in hh:mm:ss
> format.
>
> This is pretty impressive and will save substantial amount of money and
> time on larger Relion runs.
>
> Final resolution (without masking) is better with vlion too: 7.86667
> compared to 8.23256 with Relion 1.4.
>
> Cheers,
> Sargis
>
> __
> Sargis Dallakyan, PhD - Systems Administrator
> National Resource for Automated Molecular Microscopy
> Simons Electron Microscopy Center
> New York Structural Biology Center
> http://emg.nysbc.org/redmine/users/105
>
>
>
> From: Dimitry Tegunov <[log in to unmask]<mailto:[log in to unmask]>>
> To: <[log in to unmask]<mailto:[log in to unmask]>>, Sargis Dallakyan
> <[log in to unmask]<mailto:[log in to unmask]>>
> Sent: 1/25/2016 2:20 PM
> Subject: Re: Vectorized RELION
>
> Hi Sargis,
>
> looks like my system lets me get away with a mixture of C11 and C++, while
> others don't. I changed the code to use posix_memalign now. Please let me
> know if this solves the problem.
>
> Cheers,
> Dimitry
>
>
>
--
Sjors Scheres
MRC Laboratory of Molecular Biology
Francis Crick Avenue, Cambridge Biomedical Campus
Cambridge CB2 0QH, U.K.
tel: +44 (0)1223 267061
http://www2.mrc-lmb.cam.ac.uk/groups/scheres
|