Data Processing at SACLA
Obtaining metadata like detector position
1) AgBeh (silver behenate). Determine detector distance and beam center --> update SACLA-provided *.geom file (CrystFEL format) --> run sacla geom to json on *.geom to get equivalent for DIALS processing
2) h5_mpi_submit --> launches dials.stills_process with process.phil, and queueing options.
Specify what runs (integers) /work/jkern/2017B8085/xrd/r234567-0/*.h5
2.5) Data visualization.
3) metrology refinement dials.combine_experiments reference_from_experiment.detector=0.
Takes 1000 images, puts into 1 file. Output: combined_experiments.json + combined_reflections.pickle
1 Experiment = crystal + detector + beam
Must cherry pick data if there is scare data out to the corners. (but not covered here. but: largest pickle files are highly diffracting).
dials.refine combined* hierarchy_level=[0|1] # Use 0 first (refine detector as a block) then 1(refine each panel) To keep detector flat: refinement.parameterisation.detector.fix_list=Tau2,Tau3
Level 0: refine dist, shift1, shift2. Fix: tau
Level 1: refine shift1, shift2, tau1 Fix: dist, tau2, tau3
Evaluation--how do you know if it made a difference?
dev.cctbx.xfel.detector_residuals json pickle # also specify hierarchy_level=1 residuals.plot_max=0.3
program -c -e 10 -a 2# gets all config parameters for a program at expert level 10, giving all help strings.
4) redo integration with reference geometry:
reference_geometry=refined_experiments.json
5) merge
take merging script from LQ79. Take it verbatim. Use cxi.merge
- !/bin/bash
- PBS -q [smp|serial]
smp: lots of memory up to 44 pros serial: up to 14 pros, 1 node b13-occupancy: reserved for you
DIALS workflow
dials.import file.h5 (the h5 will have 1000's of images in it) --> datablock.json. Has experimental models as abstracted from image header
dials.find_spots datablock.json --> strong.pickle
dials.index strong.pickle datablock.json --> indexed.pickle experiments.json
dials.refine --> refined_experiments.json refined_reflections.pickle
dials.integrate
Aggregate processing at XFELS
Need to submit a single job for each *.h5 file (manually, or write a script) Instead of running the individual steps: dials.stills_process *.h5 process.phil
Phil file must have good parameters for data processing. Take one from previous users.
Converting SACLA pipeline geometry file to DIALS
libtbx.python modules/cctbx_project/xfel/sacla/mpccd_geom2json.py <SACLA_GEOM> distance=<DETECTOR DISTANCE>
Modifications to detector distance
Distance can be specified by modifying line 55 of /home/jkern/xfel_env/conda_install/modules/cctbx_project/dxtbx/format/FormatHDF5SaclaMPCCD.py
`self.distance = 100.0`
Submission of jobs
Use the file '/work/jkern/2017B8085/xrd/run_scripts/test_pierre.sh'
echo -n "enter phil file to use:"
read phil_inp
echo $phil_inp
for run in $(seq $1 $2);
do
for chunk in $(seq $3 $4)
do
cxi.mpi_submit input.data_dir=/work/jkern/2017B8085/xrd/data/${run}-${chunk} input.run_num=${run} \
input.dispatcher=dials.stills_process output.output_dir=/work/jkern/2017B8085/xrd/results \
input.target="/work/jkern/2017B8085/xrd/processing_phils/${phil_inp}" input.run_chunk=$chunk\
mp.method=pbs mp.queue=bl2-occupancy mp.nnodes=1 mp.nproc_per_node=28 \
mp.env_script=/home/jkern/xfel_env/setup_env.sh input.data_template="run%d-%d.h5"
done
done