1. Continuous optimization

FloatGenAlgSolver is the simplest solver (genes are plain floats, no chemistry involved) and the best place to learn the pattern every other solver follows.

The two hooks every solver needs

GenAlgSolver (the base class) runs the entire GA loop but has no idea what a chromosome means. You supply that meaning through two callables:

• assembler: turns a raw chromosome (an array of genes) into the object that gets scored. FloatGenAlgSolver’s default assembler, make_array(), is a no-op: the chromosome is the array.

• fitness_function: scores the assembled object. This is the one function you always have to write yourself.

A minimal run

from navicatGA.float_solver import FloatGenAlgSolver

def sphere(x):
    """Maximize -sum((x_i - 0.5)^2): a peak at x = [0.5, 0.5, 0.5]."""
    return -sum((xi - 0.5) ** 2 for xi in x)

solver = FloatGenAlgSolver(
    n_genes=3,
    fitness_function=sphere,
    variables_limits=(0, 1),      # shared (min, max) for every gene
    pop_size=50,
    max_gen=100,
    selection_strategy="tournament",
    random_state=42,               # reproducible run
    to_file=False,                 # skip writing output.log for this quick example
    verbose=False,
)
result = solver.solve()
print(result.best_individual, result.best_fitness)
# [0.50143372 0.50003259 0.50000241] -2.0566351926863767e-06

solve() returns a GAResult namedtuple (best_individual, best_fitness, best_pfitness, population, fitness, generations, runtime). The same values are also set on the solver as solver.best_individual_, solver.best_fitness_, etc, for code that predates GAResult.

GA-mechanics parameters (shared by every solver)

These aren’t specific to FloatGenAlgSolver: every solver forwards them to GenAlgSolver via **base_kwargs, so the same names work everywhere:

Param	Meaning
pop_size	Number of chromosomes per generation.
mutation_rate	Fraction of genes mutated per generation.
selection_rate	Fraction of the population kept as parents.
selection_strategy	roulette_wheel | tournament | two_by_two | random | boltzmann.
n_crossover_points	Number of crossover cut points.
max_gen / max_conv	Hard generation cap / generations with an unchanged best fitness before stopping early.
excluded_genes	Gene indices held fixed (never mutated/crossed).
lru_cache	Cache fitness by chromosome; needs assembler’s output to be hashable.
prune_duplicates	Drop and refill duplicate chromosomes each generation.
scalarizer	Combine multiple fitness outputs into one; see Tutorial 5.

See templates/float.yaml for the full list with defaults filled in.

variables_limits

• A single (min, max) pair applies to every gene (as above).

• A list of n_genes (min, max) pairs sets per-gene bounds, e.g. [(0, 1), (-10, 10), (0, 100)].

• Integer bounds ((0, 10) with both ints) sample integers instead of floats.

Next: Tutorial 2: SMILES fragment optimization, where the assembler actually does something.