I was looking at the
SAMIN() code in Optim (https://julianlsolvers.github.io/Optim.jl/stable/#algo/samin/) and wanted to ask if someone could help provide some intuition for the nt, ns and neps terms.
While checking some julia discourses online, there seems to be good intuition for why setting
rt to be somewhere between (0.8, 1) is a good idea, and if you want faster convergence you may want to set
rt to be low so that the algorithm can “zero in” on the optimizer quicker. In a similar vein, would setting
nt = ns = neps = 1 correspond to a more “aggressive”
SAMIN() algorithm, where by “aggressive” I mean that, at the new (lower) values of
nt, ns and neps, temperature is lowered and bounds are adjusted more quickly relative to when they are set at their default values (i.e. 5).
- What are the pitfalls of setting lower values, say 1, for nt, ns and neps?
- Are there are potential benefits of setting lower values, say 1, for nt, ns and neps (perhaps, together with a lower value of rt)?