source: branches/HeuristicLab.Problems.GrammaticalOptimization-gkr/Main/Program.cs @ 12893

using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.Globalization;
using System.Linq;
using HeuristicLab.Algorithms.Bandits;
using HeuristicLab.Algorithms.Bandits.BanditPolicies;
using HeuristicLab.Algorithms.Bandits.GrammarPolicies;
using HeuristicLab.Algorithms.GrammaticalOptimization;
using HeuristicLab.Problems.GrammaticalOptimization;

// NOTES: gkronber
// TODO: feature extraction for full symbolic expressions and experiment for all benchmark problems
// TODO: why does GaussianThompsonSampling work so well with MCTS for the artificial ant problem?
// TODO: research thompson sampling for max bandit?
// TODO: verify TA implementation using example from the original paper      
// TODO: implement thompson sampling for gaussian mixture models
// TODO: gleichzeitige modellierung von transformierter zielvariable (y, 1/y, log(y), exp(y), sqrt(y), ...)
// TODO: vergleich bei complete-randomly möglichst kurze sÀtze generieren vs. einfach zufÀllig alternativen wÀhlen
// TODO: reward discounting (fÌr verÀnderliche reward distributions Ìber zeit). speziellen unit-test dafÌr erstellen
// TODO: constant optimization 
using HeuristicLab.Problems.GrammaticalOptimization.SymbReg;
using RandomPolicy = HeuristicLab.Algorithms.Bandits.GrammarPolicies.RandomPolicy;


namespace Main {
  class Program {
    static void Main(string[] args) {
      CultureInfo.DefaultThreadCurrentCulture = CultureInfo.InvariantCulture;

      foreach (var banditPolicy in new IBanditPolicy[]
      {
        //new HeuristicLab.Algorithms.Bandits.BanditPolicies.RandomPolicy(),
        //new UCB1TunedPolicy(), 
        //new UCB1Policy(), 
        //new UCB1Policy(0.8), 
        //new UCB1Policy(1), 
        new UCB1Policy(0.5), 
        //new ExtremeHunterPolicy(), 
        //new ThresholdAscentPolicy(), 
        //new BoltzmannExplorationPolicy(1), 
        //new BoltzmannExplorationPolicy(0.5), 
        //new BoltzmannExplorationPolicy(5), 
        //new EpsGreedyPolicy(0.1), 
        //new EpsGreedyPolicy(0.05), 
      }) {
        var problem = new SymbolicRegressionPoly10Problem(500);
        var random = new Random();
        //var problem = new SymbolicRegressionProblem(random, "Vladislavleva-1", useConstantOpt: true);
        //var problem = new PrimePolynomialProblem();
        //var problem = new SantaFeAntProblem();
        var policy = new GenericGrammarPolicy(problem, banditPolicy);
        // var policy = new GenericGrammarPolicy(problem, new UCB1Policy(0.5));
        //var alg = new MonteCarloTreeSearch(problem, 23, random, new UCB1Policy(), new RandomSimulation(problem, random, 30));

        RunDemo(problem, random, policy, banditPolicy.ToString());
      }
    }


    private static void RunDemo(IProblem problem, Random random, GenericGrammarPolicy policy, string banditPolicyName) {


      for (int i = 0; i < 100; i++) {
        int iterations = 0;
        var globalStatistics = new SentenceSetStatistics();

        var alg = new SequentialSearch(problem, 23, random, 0, policy);


        alg.FoundNewBestSolution += (sentence, quality) => {
          //Console.WriteLine("{0}", globalStatistics);
        };

        alg.SolutionEvaluated += (sentence, quality) => {
          iterations++;
          globalStatistics.AddSentence(sentence, quality);
          //UpdateAlleleStatistics(sentence);
          // comment this if you don't want to see solver statistics
          if (iterations % 100 == 0) {
            if (iterations % 1000 == 0) {
              Console.Clear();
            }
            Console.SetCursorPosition(0, 0);
            Console.WriteLine("{0} {1}", iterations, string.Join(" ", policy.OptimalPulls.Take(15).Select(p => string.Format("{0:F3}", p))));
            //WriteAlleleStatistics();
            Console.WriteLine(globalStatistics.BestSentenceQuality);
            Console.WriteLine(globalStatistics.BestSentence);
            Console.WriteLine(globalStatistics);
            alg.PrintStats();
            //policy.PrintStats();
            //ResetAlleleStatistics();
          }

          // uncomment this if you want to collect statistics of the generated sentences
          //if (iterations % 1000 == 0) {
          //  Console.WriteLine("{0} {1} {2}", banditPolicyName, string.Join(" ", policy.OptimalPulls.Take(15).Select(p => string.Format("{0:F3}", p))), globalStatistics);
          //}
        };
        int maxIterations = 300000;

        // ResetAlleleStatistics();

        //var problem = new SantaFeAntProblem();
        //var problem = new RoyalPairProblem(10);
        //var problem = new FindPhrasesProblem(random, 10, 5, 3, 5, 5, 1.0, 0.9, true);
        //var problem = new PrimePolynomialProblem();
        //var problem = new SymbolicRegressionProblem(random, "Tower");
        // @"C:\reps\HeuristicLab\branches\HeuristicLab.Problems.GrammaticalOptimization\HeuristicLab.Problems.GrammaticalOptimization.SymbReg\nht-train.csv",
        //  @"C:\reps\fhooe-new\research\Datasets\Benchmark\kommenda-1.csv",
        //  1.0,
        //  true);
        // //var problem = new PrimePolynomialProblem();
        // var alg = new SequentialSearch(problem, 25, random, 0,
        //var policy = new HeuristicLab.Algorithms.Bandits.GrammarPolicies.GenericGrammarPolicy(problem, new UCB1TunedPolicy());
        //var policy = new GenericPolicy(problem);
        //var policy = new GenericGrammarPolicy(problem, new ExtremeHunterPolicy());
        //var policy = new GenericGrammarPolicy(problem, new UCB1Policy(0.5));
        //var policy = new GenericGrammarPolicy(problem, new ActiveLearningPolicy(3));
        //var policy = new GenericGrammarPolicy(problem, new IntervalEstimationPolicy());
        //var policy = new GenericGrammarPolicy(problem, new ChernoffIntervalEstimationPolicy());

        //var policy = new GenericGrammarPolicy(problem, new EpsGreedyPolicy(0.1));
        //var policy = new GenericGrammarPolicy(problem, new ExtremeHunterPolicy(0.001, 0.001, 1, 100000, minPulls: 100));
        //var policy = new GenericGrammarPolicy(problem, new ThresholdAscentPolicy(s: 1000, delta: 1));
        //var policy = new GenericGrammarPolicy(problem, new HeuristicLab.Algorithms.Bandits.BanditPolicies.UCB1TunedPolicy());
        //var policy = new GenericGrammarPolicy(problem, new HeuristicLab.Algorithms.Bandits.BanditPolicies.BoltzmannExplorationPolicy(1));
        //var policy = new GenericGrammarPolicy(problem, new HeuristicLab.Algorithms.Bandits.BanditPolicies.ThresholdAscentPolicy(500, 0.01)); // santa fe ant


        //var policy = new GenericGrammarPolicy(problem, new HeuristicLab.Algorithms.Bandits.BanditPolicies.BoltzmannExplorationWithCoolingPolicy(0.01));




        var sw = new Stopwatch();
        sw.Start();
        alg.Run(maxIterations);
        sw.Stop();

        //Console.WriteLine(globalStatistics);
        //
        //Console.WriteLine("{0:F2} sec {1,10:F1} sols/sec {2,10:F1} ns/sol",
        //  sw.Elapsed.TotalSeconds,
        //  maxIterations / (double)sw.Elapsed.TotalSeconds,
        //  (double)sw.ElapsedMilliseconds * 1000 / maxIterations);
      }
    }

    private static void UpdateAlleleStatistics(string sentence) {
      for (int i = 0; i < sentence.Length; i++) {
        var allele = sentence.Substring(i, 1);
        if (alleleStatistics.ContainsKey(allele)) alleleStatistics[allele]++;
      }
      for (int i = 0; i < sentence.Length - 2; i += 2) {
        var allele = sentence.Substring(i, 3);
        if (alleleStatistics.ContainsKey(allele)) alleleStatistics[allele]++;
      }
      for (int i = 0; i < sentence.Length - 4; i += 2) {
        var allele = sentence.Substring(i, 5);
        if (alleleStatistics.ContainsKey(allele)) alleleStatistics[allele]++;
      }
    }


    private static Dictionary<string, int> alleleStatistics;

    private static void ResetAlleleStatistics() {
      alleleStatistics = new Dictionary<string, int>()
      {
        {"a", 0},
        {"b", 0},
        {"c", 0},
        {"d", 0},
        {"e", 0},
        {"f", 0},
        {"g", 0},
        {"h", 0},
        {"i", 0},
        {"j", 0},
        {"a*b", 0},
        {"b*a", 0},
        {"c*d", 0},
        {"d*c", 0},
        {"e*f", 0},
        {"f*e", 0},
        {"a*g*i", 0},
        {"a*i*g", 0},
        {"g*a*i", 0},
        {"g*i*a", 0},
        {"i*g*a", 0},
        {"i*a*g", 0},
        {"j*c*f", 0},
        {"j*f*c", 0},
        {"c*j*f", 0},
        {"c*f*j", 0},
        {"f*c*j", 0},
        {"f*j*c", 0}
      };
    }


    private static void WriteAlleleStatistics() {
      double count = alleleStatistics.Sum(e => e.Value);
      foreach (var entry in alleleStatistics.OrderByDescending(e => e.Value)) {
        Console.WriteLine("{0,-10} {1,-10}", entry.Key, entry.Value);
      }
    }
  }
}