expanding on how best to use option selectivity 17...differentiability? #1201
Description
Adding this issue following a discussion with Jim Ianelli and Rick Methot on time-varying selectivity parameterization.
Jim: I've been struggling getting the time-varying aspect to work well [for Pollock in SS3]. One problem seems to be that the age-selectivity option 17 (with time-varying devs?) appears to have differentiation problems.
Rick: I'm not surprised that there are differentiability problems with time-varying option 17 because the age with peak selectivity will be jumping around. You may be better off with option 42, which is patterned on your pollock model approach:
#Pattern:_17; parm=nages+1; empirical as random walk N parameters to read can be overridden by setting special to non-zero
#Pattern:_41; parm=2+nages+1; // like 17, with 2 additional param for scaling (mean over bin range)
This is worth testing before the 2026 hake assessment, as well as revisiting some of the other time-varying options given recent performance.