usnistgov / MATS

Multi-spectrum Analysis Tool for Spectroscopy
https://pages.nist.gov/MATS/
Other
8 stars 7 forks source link

Fitting Speed #3

Closed eadkins333 closed 3 years ago

eadkins333 commented 3 years ago

Fitting Speed for larger files. How can we speed this up? What are the best practices for a fitting procedure.

eadkins333 commented 3 years ago

Looks like there is overhead in the Sim Model that is not related to the HTP from DF call. About 50% of the fitting time. Can we minimize this?

eadkins333 commented 3 years ago

Consider adding in convergence criteria that are less stringent to increase the speed

The args calls may let you add a tol = 1e-5 command to the minimize call, additionally you can have options = {maxiter = 2000}

eadkins333 commented 3 years ago

Got rid of the iter_cb and replaced with xtol, ftol, and iteration limits in the fit call.

eadkins333 commented 3 years ago

Check the Speed Improvements from Numba addition compared to Numba overhead

eadkins333 commented 3 years ago

Numba doesn't appear to be increasing the speed of the fits, so it was removed

eadkins333 commented 3 years ago

Re-defined the HTP definition call, so that it defines the linelist at the experimental conditions by the column instead of row by row this is a small speed up.

This issue was closed on 8/28/2019.