h52nx: improve electric current calculation.

Merged payno requested to merge improve_machine_elec_current_deduction into master


  • add an option to skip this part
    • test it on real dataset
  • speed up the interpolation
    • move calculation to multi-thread
      • parallelize research of time stamp and machine electrical current interpolation which are the part taking most of the time.
    • reduce research time
      • both timestamp and know_timestamp are ordered: only look forward (add start_index).
        • force to filter research timestamp before but algorithmic gain is substantial (and remove workaround we did to speed up)
      • filter know_timestamp (remove energies surround by the same energy ...) -> seems to reduce the known_electric_current about a factor of 4 to 5. Still good to take (speed up time stamp research)
    • test it on real dataset

dummy figures

on the datasets (180k projections) I tested (from ludo / ID19) I had to stop this processing because it was taking way to much time (>10 minutes). I think it was way more. Anyway such amount of time is unacceptable for moving to a NXtomo.

With this improvement it takes ~~ ~ 10s on the gpu partition with (cpu_count() == 64). (note: 'only' half of the available cpu are used).~~ 4s in on gpu since using numpy.searchsorted thanks to @mirone


result on the dataset processed

image image on this last one the lost of the beam was 'real' and expected...

Edited by payno

Merge request reports