Closed emilycollier closed 3 years ago
Also, scipy.optimize.newton is even faster (+50% in my tests) with minimal impact on solution (cumulative MB after 12 days differs by less than 0.5% for all three surface temperature solutions) although there is no lower bound imposed. Let me know if you would like a PR for these options.
Added both SLSQP method and scipy.optimize.newton (secant method) as options in new release on master (COSIPY v1.4).
My tests show surface temperature update is 25% faster with local bounded method 'SLSQP' compared with ’L-BFGS-B’ with minimal impact on solution for both the log and adaptive re-meshing options.