Я пытаюсь сделать символическую регрессию геометрической модели.И в большинстве случаев он застревает с показателем физической подготовки, который не близок к 0. Поэтому я провел пару исследований и выяснил, что это проблема локальных минимумов.И некоторые люди пытались расставить приоритеты разнообразия населения над фитнесом.Но это не то, что я хочу.
Так что я решил перенастроить algorithms.eaSimple и добавил в него блок.Таким образом, это сбрасывает население, когда последнее поколение n=50 имеет такую же пригодность.Я понятия не имею, кроме этого, поскольку я очень новичок в этом. Есть ли лучший способ сделать это?
Я использую минимальную пригодность.creator.create("FitnessMin", base.Fitness, weights=(-1.0,))
def my_eaSimple(population, toolbox, cxpb, mutpb, ngen, stats=None, halloffame: tools.HallOfFame = None,
verbose=True):
logbook = tools.Logbook()
logbook.header = ['gen', 'nevals'] + (stats.fields if stats else [])
# Evaluate the individuals with an invalid fitness
invalid_ind = [ind for ind in population if not ind.fitness.valid]
fitnesses = toolbox.map(toolbox.evaluate, invalid_ind)
for ind, fit in zip(invalid_ind, fitnesses):
ind.fitness.values = fit
if halloffame is not None:
halloffame.update(population)
record = stats.compile(population) if stats else {}
logbook.record(gen=0, nevals=len(invalid_ind), **record)
if verbose:
print(logbook.stream)
# Begin the generational process
gen = 1
last_few_pop_to_consider = 50
starting_condition = last_few_pop_to_consider
is_last_few_fitness_same = lambda stats_array: abs(numpy.mean(stats_array) - stats_array[0]) < 0.1
while gen < ngen + 1:
# Select the next generation individuals
offspring = toolbox.select(population, len(population))
# Vary the pool of individuals
offspring = algorithms.varAnd(offspring, toolbox, cxpb, mutpb)
# Evaluate the individuals with an invalid fitness
invalid_ind = [ind for ind in offspring if not ind.fitness.valid]
fitnesses = toolbox.map(toolbox.evaluate, invalid_ind)
for ind, fit in zip(invalid_ind, fitnesses):
ind.fitness.values = fit
# Update the hall of fame with the generated individuals
if halloffame is not None:
halloffame.update(offspring)
# Replace the current population by the offspring
population[:] = offspring
# Append the current generation statistics to the logbook
record = stats.compile(population) if stats else {}
logbook.record(gen=gen, nevals=len(invalid_ind), **record)
if verbose:
print(logbook.stream)
gen += 1
# stopping criteria
min_fitness = record['fitness']['min\t']
# max_fitness = record['fitness']['max\t']
if min_fitness < 0.1:
print('Reached desired fitness')
break
if gen > starting_condition:
min_stats = logbook.chapters['fitness'].select('min\t')[-last_few_pop_to_consider:]
if is_last_few_fitness_same(min_stats):
print('Defining new population')
population = toolbox.population(n=500)
starting_condition = gen + last_few_pop_to_consider
return population, logbook
Выходные данные
gen nevals avg max min std avg max min std
0 500 2.86566e+23 1.41421e+26 112.825 6.31856e+24 10.898 38 3 9.50282
1 451 2.82914e+18 1.41421e+21 90.113 6.31822e+19 6.226 38 1 5.63231
2 458 2.84849e+18 1.41421e+21 89.1206 6.3183e+19 5.602 36 1 5.18417
3 459 4.24902e+14 2.01509e+17 75.1408 9.01321e+15 5.456 35 1 4.05167
4 463 4.23166e+14 2.03904e+17 74.3624 9.11548e+15 6.604 36 1 3.61762
5 462 2.8693e+11 1.25158e+14 65.9366 5.60408e+12 7.464 34 1 3.00478
6 467 2.82843e+18 1.41421e+21 65.9366 6.31823e+19 8.144 37 1 3.51216
7 463 5.40289e+13 2.65992e+16 65.9366 1.1884e+15 8.322 22 1 2.88276
8 450 6.59849e+14 3.29754e+17 59.1286 1.47323e+16 8.744 34 1 3.03685
9 458 1.8128e+11 8.17261e+13 54.4395 3.65075e+12 9.148 23 1 2.69557
10 459 6.59851e+14 3.29754e+17 54.4395 1.47323e+16 9.724 35 1 3.02255
11 458 2.34825e+10 1.41421e+11 54.4395 5.26173e+10 9.842 18 1 2.32057
12 459 3.52996e+11 1.60442e+14 54.4395 7.1693e+12 10.56 33 1 2.63788
13 457 3.81044e+11 1.60442e+14 54.4395 7.18851e+12 11.306 35 1 2.84611
14 457 2.30681e+13 1.15217e+16 54.4395 5.14751e+14 11.724 24 1 2.6495
15 463 2.65947e+10 1.41421e+11 54.4395 5.52515e+10 12.072 29 1 2.63036
16 469 4.54286e+10 9.13693e+12 54.4395 4.10784e+11 12.104 34 1 3.00752
17 461 6.58255e+11 1.74848e+14 54.4395 9.76474e+12 12.738 36 4 3.10956
18 450 2.03669e+10 1.41421e+11 54.4395 4.96374e+10 13.062 30 4 3.01963
19 465 1.75385e+10 2.82843e+11 54.4395 4.74595e+10 13.356 24 1 2.82157
20 458 1.83887e+10 1.41421e+11 54.4395 4.7559e+10 13.282 23 1 3.03949
21 455 3.67899e+10 8.36173e+12 54.4395 4.04044e+11 13.284 34 4 3.03106
22 461 1.36372e+10 1.41422e+11 54.4395 4.16569e+10 13.06 35 3 3.01005
23 471 2.00634e+26 1.00317e+29 54.3658 4.48181e+27 12.798 36 1 3.17698
24 466 2.82843e+18 1.41421e+21 54.3658 6.31823e+19 12.706 36 3 3.07043
25 464 3.00384e+10 8.36174e+12 54.3658 3.75254e+11 12.612 34 5 2.89231
26 474 2.00925e+10 1.41421e+11 54.3658 4.93588e+10 12.594 34 3 2.60253
27 452 2.9528e+11 1.41626e+14 54.3658 6.32694e+12 12.43 25 1 2.49822
28 453 1.23899e+10 1.41421e+11 54.3658 3.98511e+10 12.41 20 5 2.45721
29 456 5.98529e+14 2.99256e+17 54.3658 1.33697e+16 12.57 37 1 2.6346
30 474 1.35672e+13 6.69898e+15 54.3658 2.99297e+14 12.526 35 1 2.94029
31 446 6.92755e+22 3.46377e+25 54.3658 1.5475e+24 12.55 36 1 2.62517
32 462 4.02525e+10 8.16482e+12 54.3658 3.92769e+11 12.764 34 5 2.77061
33 449 1.53268e+13 7.65519e+15 54.3658 3.42007e+14 12.628 35 1 2.76218
34 466 3.13214e+16 1.54388e+19 54.3658 6.89799e+17 12.626 35 1 2.97626
35 464 2.82845e+18 1.41421e+21 54.3658 6.31823e+19 12.806 36 5 2.74597
36 460 2.93493e+11 1.32308e+14 54.3658 5.91505e+12 12.734 35 5 2.88084
37 456 2.93491e+10 8.29826e+12 54.3658 3.72372e+11 12.614 37 1 2.80517
38 449 3.44519e+10 8.16482e+12 54.3658 3.67344e+11 12.742 34 3 2.91881
39 466 1.53217e+13 7.65519e+15 54.3658 3.42008e+14 12.502 35 3 2.70296
40 454 2.82843e+18 1.41421e+21 54.3658 6.31823e+19 12.51 36 1 2.81103
41 453 9.66059e+24 4.68888e+27 54.3658 2.09566e+26 12.554 33 1 2.47691
42 448 2.2287e+10 3.38289e+12 54.3658 1.58629e+11 12.576 26 1 2.50763
43 460 5.47399e+12 2.73042e+15 54.3658 1.21985e+14 12.584 34 1 2.80053
44 460 2.82843e+18 1.41421e+21 54.3658 6.31823e+19 12.692 27 1 2.86516
45 464 2.829e+18 1.41421e+21 54.3658 6.31823e+19 12.57 34 1 3.15549
46 460 2.92607e+11 1.31556e+14 54.3658 5.88776e+12 12.61 37 3 2.78817
47 465 2.82843e+18 1.41421e+21 54.3658 6.31823e+19 12.622 36 1 3.04616
48 461 1.64306e+10 2.97245e+12 54.3658 1.37408e+11 12.468 26 1 2.57856
49 463 1.54834e+10 1.41421e+11 54.3658 4.4029e+10 12.464 20 1 2.4529
50 451 1.59239e+10 1.41421e+11 54.3658 4.44609e+10 12.63 33 1 2.76281
51 455 5.40036e+19 2.70018e+22 54.3658 1.20635e+21 12.78 37 1 2.84668
52 478 2.82843e+18 1.41421e+21 54.3658 6.31823e+19 12.712 36 3 2.84694
53 461 2.78669e+21 1.39193e+24 54.3658 6.21866e+22 12.714 36 1 3.23546
54 471 7.41272e+12 3.70045e+15 54.3658 1.65323e+14 12.336 34 3 2.848
55 465 2.83036e+18 1.41421e+21 54.3658 6.31822e+19 12.74 36 1 3.62662
56 459 2.82843e+18 1.41421e+21 54.3658 6.31823e+19 12.606 29 1 2.60437
57 453 5.98308e+24 2.99154e+27 54.3658 1.33652e+26 12.722 34 1 2.62311
58 460 3.62463e+21 1.8109e+24 54.3658 8.09047e+22 12.65 37 1 2.92361
Defining new population
59 500 5.83025e+48 2.91513e+51 109.953 1.30238e+50 10.846 38 1 8.89889
60 464 2.93632e+15 8.87105e+17 165.988 4.38882e+16 5.778 36 1 4.79173
61 444 5.54852e+19 2.70018e+22 93.5182 1.20674e+21 4.992 37 1 4.648
62 463 4.28647e+14 2.14148e+17 82.0774 9.56741e+15 5.468 34 1 4.34891
63 464 2.82843e+18 1.41421e+21 78.8184 6.31823e+19 6.624 35 1 4.25989
64 453 3.40035e+11 1.60954e+14 68.7629 7.19022e+12 7.356 36 1 3.77694
65 456 5.65762e+18 2.82851e+21 68.7629 1.26368e+20 7.606 35 1 4.15966
66 461 2.82843e+18 1.41421e+21 68.7629 6.31823e+19 7.906 35 1 3.81171
67 447 1.63302e+10 1.41421e+11 68.7629 4.51102e+10 7.802 33 1 3.47258
68 463 6.59552e+14 3.29754e+17 68.7629 1.47323e+16 8.37 34 3 3.80698
69 460 1.53579e+13 7.65512e+15 68.7629 3.42003e+14 8.646 35 1 3.64042
70 461 2.80014e+10 1.41421e+11 68.7629 5.63553e+10 9.212 38 1 3.69582
71 453 1.97446e+11 7.80484e+13 68.7629 3.50764e+12 9.84 34 1 3.74785
72 459 9.98853e+11 1.75397e+14 68.7629 1.25317e+13 10.284 35 3 3.61764
73 453 5.6863e+16 2.84218e+19 68.7629 1.26979e+18 10.796 36 1 3.86864
74 466 2.57445e+10 1.41434e+11 68.7629 5.4564e+10 10.806 35 1 3.2949
75 453 2.82849e+18 1.41421e+21 68.7629 6.31823e+19 10.876 34 1 3.27301
76 433 1.67235e+20 8.36174e+22 68.7629 3.73574e+21 10.868 35 1 2.94051
77 457 3.6663e+21 1.83315e+24 68.7629 8.1899e+22 10.964 37 3 3.21476
78 461 1.80829e+14 9.04015e+16 68.7629 4.03883e+15 10.992 35 3 3.26985
79 450 3.21984e+11 1.41626e+14 68.7629 6.32593e+12 11.17 28 1 2.77941
80 460 2.82843e+18 1.41421e+21 68.7629 6.31823e+19 11.044 35 1 3.25362
81 455 6.46751e+14 2.99308e+17 68.7629 1.34123e+16 11.06 34 1 3.51061
82 463 3.21908e+21 1.60954e+24 68.7629 7.19088e+22 11.112 34 1 3.58433
83 473 2.82843e+18 1.41421e+21 68.7629 6.31823e+19 10.946 38 3 3.70663
84 460 3.14081e+11 1.41626e+14 68.7629 6.32625e+12 10.896 35 1 3.4976
85 456 1.53419e+13 7.65526e+15 68.7629 3.4201e+14 11.156 36 1 3.23661
Популяция сбрасывается после получения 54,4395 минимального уровня пригодности для 50 раз в 59-м поколении.