sonar_tutorial_part3

How to find a better candidate solution?

Brute force approach

Tweaking the search parameters is the usual first approach.

Increasing the search duration (more generations), running more iterations, and enlarging the population size are some ways to balance time and quality (this can be done via the parameters class):

int main()
{
  using namespace ultra;

  src::dataframe::params params;
  params.output_index = src::dataframe::params::index::back;

  src::problem prob("sonar.csv", params);
  prob.setup_symbols();

  // TWEAKING THE PARAMETERS
  prob.params.population.init_subgroups =    3;  // <-- 1
  prob.params.population.individuals    = 3000;  // <-- 2
  prob.params.evolution.generations     =  200;  // <-- 3

  src::search s(prob);
  s.validation_strategy<src::holdout_validation>(prob);

  const auto result(s.run(5));                   // <-- 4

  std::cout << "\nCANDIDATE SOLUTION\n"
            << out::c_language << result.best_individual
            << "\n\nACCURACY\n" << *result.best_measurements.accuracy * 100.0
            << '%'
            << "\n\nFITNESS\n" << *result.best_measurements.fitness << '\n';
}

(the above code is in the examples/sonar03.cc file)

1. prob.params.population.init_subgroups = 3;

   The search starts with 3 distinct sub-populations evolving in parallel. Ultra uses an algorithm called ALPS and can take advantage of multithreading / multiprocessing architecture.

2. prob.params.population.individuals = 3000;

   A larger population permits a broader exploration of the search space (large populations require more computing power / memory).

3. prob.params.evolution.generations = 200;

   Allowing more time to refine a candidate solution increases the likelihood of improvement but may also lead to resource wastage in cases of premature convergence.

4. s.run(5)

   We want a search consisting of five runs, each representing an independent evolution cycle. The solution of each run is compared against the validation set, and the best candidate solution is chosen as the winner.

Evolution customization

(full example in examples/sonar04.cc)

Changing some aspects of the evolution process can be beneficial. Two interesting changes can be:

• prob.params.evolution.brood_recombination = 3. Brood recombination crossover is an effective technique for exploring the potential of the crossover operator in greater depth. While it may occasionally lead to premature convergence, it's worth experimenting with.
• prob.params.team.individuals = 3. The cooperative team approach results in an improved training and generalization performance compared to the standard GP. However, caution is necessary even in this scenario: an excessively large team may lead to overfitting.

[INFO] Importing dataset...
[INFO] ...dataset imported
[INFO] Examples: 208, features: 60, classes: 2
[INFO] Setting up terminals...
[INFO] Category 0 variables: `X1` `X2` `X3` `X4` `X5` `X6` `X7` `X8` `X9` `X10` `X11` `X12` `X13` `X14` `X15` `X16` `X17` `X18` `X19` `X20` `X21` `X22` `X23` `X24` `X25` `X26` `X27` `X28` `X29` `X30` `X31` `X32` `X33` `X34` `X35` `X36` `X37` `X38` `X39` `X40` `X41` `X42` `X43` `X44` `X45` `X46` `X47` `X48` `X49` `X50` `X51` `X52` `X53` `X54` `X55` `X56` `X57` `X58` `X59` `X60`
[INFO] ...terminals ready
[INFO] Automatically setting up symbol set...
[INFO] Category 0 symbols: `FABS` `FADD` `FDIV` `FLN` `FMUL` `FMOD` `FSUB`
[INFO] ...symbol set ready
[INFO] Holdout validation settings: 70% training (144), 30% validation (64), 0% test (0)
[INFO] Number of layers set to 1
[INFO] Population size set to 736
   0.592       0:     -43.8549
   1.063       1:     -39.3506
   1.520       2:     -35.7575

...

   03:04     386:     -19.8824
   03:06     390:      -19.605
   03:07     392:     -19.3401
   03:08     393:     -19.0378
   03:10     396:      -18.913
   03:13     401:     -18.1754
   03:17     407:     -18.1338
   03:18     408:     -18.1247
   03:20     411:     -17.9289
   03:23     415:     -17.9113
   05:11     578:     -17.5204
   05:17     586:     -17.4367
   05:19     589:     -17.1784
   05:22     593:     -17.1744
   05:22     594:     -17.1669
[INFO] Evolution completed at generation: 602. Elapsed time: 05:27
Run 4 TRAINING. Fitness: -17.1669  Accuracy: 93.055556%
Run 4 VALIDATION. Fitness: -19.4064  Accuracy: 75.000000%

CANDIDATE SOLUTION
X11+fmod(X47,(X28-X15))
fmod(X42,(X19*X18))
(X36+X31)*X17


ACCURACY
89.0625%

PROCEED TO PART 4 →