Performance test case for x-ray spectral fitting

see generate.py (for sherpa)

see generatexspec.py (for xspec)

Benchmark

How fast is fitting if only the normalisations are varied?

This should be trivially fast (<10ms), as it is only matrix-vector multiplication, and the poisson formula.

Model setup

• 3 spectra (Chandra + 2x NuSTAR)

• model

  • sum of 8 apecs
  • table models (download from https://zenodo.org/record/1169181)

• statistic: Poisson (cstat)

• backgrounds

  • in sherpa: model set (powerlaw)
  • in xspec: no model (WStat)

Current results

Sherpa:

• 14.89 model evaluations per second
• 21 model evaluations per second with scale1d

Xspec:

• 2 model evaluations per second