Motivation
Development of this application was stimulated by the discussion with Marcin Sikorski, doing xcs experiments.
Users need in real-time algorithm for calculation of image vs time auto-correlation function
g2(tau) = <I(t)*I(t+tau)> / (<I(t)> * <I(t+tau)>),
where I(t) is an intensity at time t, and tau is a delay between two measurements.
Typical experimental condition can be described as follows:
- Run duration is about one hour at frequency up to 120 Hz that gives up to 500-10^6 images.
- Currently typical imaging devise is a Princeton camera with 1300x1340 pixels.
- Need to calculate
g2(tau)
for each pixel, averaged over all possible image timest
with time differencetau
between images. - A set of
tau
should have about 30-100 points in log scale uniformly covering the run duration. - Use for example xcsi0112-r0015: 500 images with 8 sec between images.
Desired time for evaluation of the auto-correlation function should be comparable with run duration ~<1 hour. Currently this algorithm takes a few hours that can not be used for fast feedback in real time experiment.
Algorithm
Basic idea is (1) to split image vs time for small parts in image, (2) to process each part on separate computer node, (3) to merge results at the end of processing. It is clear that significant speedup (~T/Nnodes) is achieved at the 2nd stage. These three stages are performed in separate C++ applications. Wrapping python script allows to submit job by a single command. It takes care about numerous files and sub-process management in this job, as discussed below.
Code location
All modules for this application resides in the package ImgAlgos:
Module |
Functionality |
---|---|
ImgVsTimeSplitInFiles |
splitter |
CorAna |
base class with common methods |
CorAnaData |
data processing for split files |
CorAnaInputParameters |
provides storage for input parameters |
CorAnaMergeFiles |
merging algorithm |
CorAnaPars.py |
singleton class for parameter storage in the wrapping file manager |
CorAnaSubmit.py |
global methods for the file manager |
app/corana_submit |
sequence of procedures in the file manager |
Image splitting
Image splitting is implemented as a regular psana module ImgAlgos::ImgVsTimeSplitInFiles.
Command:
psana -c <config-file> <xtc-file-list>
For example:
psana -c ana-misc-exp/psana-xcsi0112-r0015-img-auto-correlation.cfg /reg/d/psdm/XCS/xcsi0112/xtc/e167-r0015-*
Produce files:
img-xcs-r0015-b0000.bin - file with a part of image vs time img-xcs-r0015-b0001.bin img-xcs-r0015-b0002.bin img-xcs-r0015-b0003.bin img-xcs-r0015-b0004.bin img-xcs-r0015-b0005.bin img-xcs-r0015-b0006.bin img-xcs-r0015-b0007.bin img-xcs-r0015-time.txt - list of time-record for all events in processed run. img-xcs-r0015-time-ind.txt - list of time-record for all events in processed run with time index. img-xcs-r0015-med.txt - file with metadata. In particular it has the original image size, number of image parts for splitting, number of images in run, etc.
Algorithms:
- The <int16_t> image data array is split for ordered number of equal parts and each part is saved in the output
img-xcs-r0015-b000N.bin
file sequentially for all selected events. - The time record for each selected event is saved in the file
img-xcs-r0015-time.txt
. - At the end of the splitting procedure:
- the average time difference and its rms between sequential events is evaluated for all recorded time records.
- The file
img-xcs-r0015-time.txt
is re-winded and for each record the time index is evaluated as unsigned value of<time-index> = (<event-time> + 0.5 <average-time-between-events>) / <average-time-between-events>
- Event record with time index is saved in the file
img-xcs-r0015-time.txt
- The file
img-xcs-r0015-med.txt
contains input parameters, image size,<average-time-between-events
, maximal value of the time index etc.
Time correlation processing
ImgAlgos/test/corana
application
Command:
<path>/corana -f <fname-data> [-t <fname-tau>] -h [-l <logfile>] [-b <basedir>]
For example the interactive and batch mode commands:
<path>/corana -f img-xcs-r0015-b0001.bin -t my-tau.txt bsub -q psnehq -o ~/<path-from-home-dir>/log.txt '<path>/corana -f img-xcs-r0015-b0000.bin'
Where ~/<path-from-home-dir>/
or <path>
is arbitrary choice and is a matter of taste.
Produce files:
img-xcs-r0015-tau.txt - string of tau values for which the auto-correlation function is evaluated img-xcs-r0015-b0000-result.bin - auto-correlators for the part of the image for all tau values img-xcs-r0015-b0001-result.bin img-xcs-r0015-b0002-result.bin img-xcs-r0015-b0003-result.bin img-xcs-r0015-b0004-result.bin img-xcs-r0015-b0005-result.bin img-xcs-r0015-b0006-result.bin img-xcs-r0015-b0007-result.bin
Merging results
ImgAlgos/test/corana_merge
application
Command:
<path>/corana_merge -f <fname-data> [-t <fname-tau>] -h [-l <logfile>] [-b <basedir>]
For example:
<path>/corana_merge -f img-xcs-r0015-b0001-result.bin -t my-tau.txt
Produce file:
img-xcs-r0015-image-result.bin
Wrapping script
./CorAnaSubmit.py
./CorAnaSubmit.py -c <config-file> -t <fname-tau> [-x] <xtc-file-list>
For example:
./CorAnaSubmit.py -c ana-misc-exp/psana-xcsi0112-r0015-img-auto-correlation.cfg -t my-tau.txt /reg/d/psdm/XCS/xcsi0112/xtc/e167-r0015-s00-c00.xtc
File formats
img-xcs-r0015-b000N.bin
:
Currently it saves<uint16_t>
amplitude for each pixel in binary format for:<data-for-img-partN-of-img1> <data-for-img-partN-of-img2> ... <data-for-img-partN-of-imgLast>
img-xcs-r0015-med.txt
:IMAGE_ROWS 1300 IMAGE_COLS 1340 IMAGE_SIZE 1742000 NUMBER_OF_FILES 8 BLOCK_SIZE 217750 REST_SIZE 0 NUMBER_OF_IMGS 500 FILE_TYPE bin DATA_TYPE uint16_t TIME_SEC_AVE 8.088413 TIME_SEC_RMS 0.063639 TIME_INDEX_MAX 499
img-xcs-r0015-time.txt
:where each record has:1 0.000000 0.000000 20120616-080236.671607864 5366 0 2 8.026429 8.026429 20120616-080244.698036743 8255 1 3 16.144788 8.118359 20120616-080252.816395836 11177 2 4 24.154835 8.010048 20120616-080300.826443448 14060 3 ...
<image-in-file#> <t(sec)-from-the-1st-event> <dt(sec)> <time-stamp> <fiducials> <event#-since-configure>
img-xcs-r0015-time.txt
:where each record has:1 0.000000 0.000000 20120616-080236.671607864 5366 0 0 2 8.026429 8.026429 20120616-080244.698036743 8255 1 1 3 16.144788 8.118359 20120616-080252.816395836 11177 2 2 4 24.154835 8.010048 20120616-080300.826443448 14060 3 3 5 32.281937 8.127102 20120616-080308.953545010 16985 4 4 ...
<image-in-file#> <t(sec)-from-the-1st-event> <dt(sec)> <time-stamp> <fiducials> <event#-since-configure> <time-index-starting-from-0>
img-xcs-r0015-b000N-result.bin
:
Currently it saves<float>
correlator for each pixel in binary format for:<corr-for-img-partN-of-tau1> <corr-for-img-partN-of-tau2> ... <corr-for-img-partN-of-tauLast>
my-tau.txt
:contains the tau values presented in terms of number of ordered images in the file.1 3 5 7 9 10 12 14 16 18 20 24 28 30 32 36 40 ... 160 180 200 240 280 300 320 360 400