these things are really important! and could well be super broken. they should be tested

https://github.com/grantjenks/sorted_containers

try runs with these guys commented and uncommented. Could be a lot faster, given the number of accesses in the loops.

like here. What we want is for the functions in the list to have __name__s. Maybe provide an api for it. But just attaching them to the model is dumb.

There should be a 'state' object passed around during search, holding

• input vector
• output vector
• total energy

for

• the best
• the current state

Low-priority, but: is

    A = sum(a_matrix[i][j] * math.sin(alpha[j]) +
            b_matrix[i][j] * math.cos(alpha[j])
            for i in xrange(d) for j in xrange(d))

faster or slower than iterating over the ranges in two sums, then summing the sums? There would be a few more generators, but far fewer additions.

It probably doesn't matter at this scale, and the current implementation is more maintainable. But, I'm curious.

so something like

sc = SearcherConfig(MaxWalkSat, {'cooling_rate': .3})

and then to construct you could do

s = sc(model=ZDT1)

so all args could be passed at configuration time or at construction time.

Is there any reason not to use Counter for SymbolLog?

(like in GA 👍)