Early stopping ends iterative optimization before its configured limit when a target condition is met. The condition may be maximum fitness, acceptable accuracy, or a chosen loss value. section 3.2 section 5.1.1
A practical early-stopping workflow:
- Set a deliberately generous generation or epoch limit. section 5.1.1
- Measure the relevant result after evaluation, such as fitness, accuracy, or loss. section 3.2 section 6.2
- Break the loop as soon as the configured target is reached, optionally saving weights for later continuation. section 5.1.1 section 6.2
Queen placement with DEAP
A genetic algorithm evolves a population of queen placements with a maximum-fitness target equal to the number of queens. Although 100 generations are allowed, the run reaches a valid solution after 67 generations and stops. section 3.2
MLP optimization
A genetic algorithm evaluates neural-network weights on a circles dataset and stops when accuracy reaches an acceptable value, rather than completing every planned generation. section 6.2
Listing 3.10 EDL_3_2_QueensGambit.ipynb: The evolution function
for ind, fit in zip(invalid_ind, fitnesses): #1
ind.fitness.values = fit
if fit[0] >= max: #2
print("Solved") #3
done = True #3
Inspect the minimal evolution-loop condition that marks the run complete when an individual's fitness reaches the maximum. section 3.2
