using System;
using System.Collections.Generic;
using System.Globalization;
using System.Linq;
using System.Text;
using System.Threading.Tasks;
using HeuristicLab.Algorithms.Bandits;
using HeuristicLab.Algorithms.Bandits.BanditPolicies;
using HeuristicLab.Algorithms.Bandits.GrammarPolicies;
using HeuristicLab.Algorithms.Bandits.Models;
using HeuristicLab.Algorithms.GrammaticalOptimization;
using HeuristicLab.Common;
using HeuristicLab.Problems.GrammaticalOptimization.SymbReg;
using Microsoft.VisualStudio.TestTools.UnitTesting;
using RandomPolicy = HeuristicLab.Algorithms.Bandits.BanditPolicies.RandomPolicy;

namespace HeuristicLab.Problems.GrammaticalOptimization.Test {
  [TestClass]
  public class TestTunedSettings {
    private const int randSeed = 31415;
    internal class Configuration {
      public IProblem Problem;
      public int MaxSize;
      public int RandSeed;

      public override string ToString() {
        return string.Format("{0} {1} {2}", RandSeed, Problem, MaxSize);
      }
    }
    [TestMethod]
    [Timeout(1000 * 60 * 60 * 12)] // 12 hours
    // this configuration worked especially well in the experiments
    public void TestAllPoliciesArtificialAnt() {
      CultureInfo.DefaultThreadCurrentCulture = CultureInfo.InvariantCulture;

      var instanceFactories = new Func<int, ISymbolicExpressionTreeProblem>[]
      {
        (randSeed) => (ISymbolicExpressionTreeProblem)new SantaFeAntProblem(), 
      };

      var policyFactories = new Func<IBanditPolicy>[]
        {
         () => new RandomPolicy(), 
         () => new ActiveLearningPolicy(),  
         () => new EpsGreedyPolicy(0.01, (aInfo)=> aInfo.MaxReward, "max"), 
         () => new EpsGreedyPolicy(0.05, (aInfo)=> aInfo.MaxReward, "max"), 
         () => new EpsGreedyPolicy(0.1, (aInfo)=> aInfo.MaxReward, "max"), 
         () => new EpsGreedyPolicy(0.2, (aInfo)=> aInfo.MaxReward, "max"), 
         //() => new GaussianThompsonSamplingPolicy(), 
         () => new GaussianThompsonSamplingPolicy(true), 
         () => new GenericThompsonSamplingPolicy(new GaussianModel(0.5, 10, 1)), 
         () => new GenericThompsonSamplingPolicy(new GaussianModel(0.5, 10, 1, 1)), 
         //() => new BernoulliThompsonSamplingPolicy(),
         () => new GenericThompsonSamplingPolicy(new BernoulliModel(1, 1)), 
         () => new EpsGreedyPolicy(0.01), 
         () => new EpsGreedyPolicy(0.05), 
         () => new EpsGreedyPolicy(0.1), 
         () => new EpsGreedyPolicy(0.2), 
         () => new EpsGreedyPolicy(0.5), 
         () => new UCTPolicy(0.01),
         () => new UCTPolicy(0.05),
         () => new UCTPolicy(0.1),
         () => new UCTPolicy(0.5),
         () => new UCTPolicy(1),
         () => new UCTPolicy(2),
         () => new UCTPolicy( 5),
         () => new UCTPolicy( 10),
         () => new ModifiedUCTPolicy(0.01),
         () => new ModifiedUCTPolicy(0.05),
         () => new ModifiedUCTPolicy(0.1),
         () => new ModifiedUCTPolicy(0.5),
         () => new ModifiedUCTPolicy(1),
         () => new ModifiedUCTPolicy(2),
         () => new ModifiedUCTPolicy( 5),
         () => new ModifiedUCTPolicy( 10),
         () => new UCB1Policy(), 
         () => new UCB1TunedPolicy(), 
         () => new UCBNormalPolicy(), 
         () => new BoltzmannExplorationPolicy(1),
         () => new BoltzmannExplorationPolicy(10),
         () => new BoltzmannExplorationPolicy(20),
         () => new BoltzmannExplorationPolicy(100),
         () => new BoltzmannExplorationPolicy(200),
         () => new BoltzmannExplorationPolicy(500),
         () => new ChernoffIntervalEstimationPolicy( 0.01), 
         () => new ChernoffIntervalEstimationPolicy( 0.05),
         () => new ChernoffIntervalEstimationPolicy( 0.1),
         () => new ChernoffIntervalEstimationPolicy( 0.2),
         () => new ThresholdAscentPolicy(5, 0.01), 
         () => new ThresholdAscentPolicy(5, 0.05),
         () => new ThresholdAscentPolicy(5, 0.1),
         () => new ThresholdAscentPolicy(5, 0.2),
         () => new ThresholdAscentPolicy(10, 0.01), 
         () => new ThresholdAscentPolicy(10, 0.05),
         () => new ThresholdAscentPolicy(10, 0.1),
         () => new ThresholdAscentPolicy(10, 0.2),
         () => new ThresholdAscentPolicy(50, 0.01), 
         () => new ThresholdAscentPolicy(50, 0.05),
         () => new ThresholdAscentPolicy(50, 0.1),
         () => new ThresholdAscentPolicy(50, 0.2),
         () => new ThresholdAscentPolicy(100, 0.01), 
         () => new ThresholdAscentPolicy(100, 0.05),
         () => new ThresholdAscentPolicy(100, 0.1),
         () => new ThresholdAscentPolicy(100, 0.2),
         () => new ThresholdAscentPolicy(500, 0.01), 
         () => new ThresholdAscentPolicy(500, 0.05),
         () => new ThresholdAscentPolicy(500, 0.1),
         () => new ThresholdAscentPolicy(500, 0.2),
         () => new ThresholdAscentPolicy(5000, 0.01), 
         () => new ThresholdAscentPolicy(10000, 0.01), 
        };
      var maxSizes = new int[] { 17 };  // necessary size for ant programm
      int nReps = 20;
      int maxIterations = 100000;
      foreach (var instanceFactory in instanceFactories) {
        var sumBestQ = 0.0;
        var sumItersToBest = 0;
        double fractionSolved = 0.0;
        foreach (var conf in GenerateConfigurations(instanceFactory, nReps, maxSizes)) {
          foreach (var policy in policyFactories) {
            var prob = conf.Problem;
            var maxLen = conf.MaxSize;
            var rand = new Random(conf.RandSeed);

            var solver = new SequentialSearch(prob, maxLen, rand, 0, new GenericGrammarPolicy(prob, policy(), true));
            var problemName = prob.GetType().Name;
            var policyName = policy().ToString();
            double bestQ; int itersToBest;
            RunSolver(solver, problemName, policyName, 1.0, maxIterations, maxLen, out bestQ, out itersToBest);
            sumBestQ += bestQ;
            sumItersToBest += itersToBest;
            if (bestQ.IsAlmost(1.0)) fractionSolved += 1.0 / nReps;
          }
        }
        // Assert.AreEqual(0.85, fractionSolved, 1E-6);
        // Assert.AreEqual(0.99438202247191, sumBestQ / nReps, 1E-6);
        // Assert.AreEqual(5461.7, sumItersToBest / (double)nReps, 1E-6);
      }
    }


    [TestMethod]
    [Timeout(1000 * 60 * 60 * 12)] // 12 hours
    // this configuration worked especially well in the experiments
    public void TestPoly10WithOutConstantOpt() {
      CultureInfo.DefaultThreadCurrentCulture = CultureInfo.InvariantCulture;

      var instanceFactories = new Func<int, ISymbolicExpressionTreeProblem>[]
      {
        (randSeed) => (ISymbolicExpressionTreeProblem)new SymbolicRegressionPoly10Problem(), 
      };

      var maxSizes = new int[] { 23 };  
      int nReps = 20;
      int maxIterations = 100000;
      foreach (var instanceFactory in instanceFactories) {
        var sumBestQ = 0.0;
        var sumItersToBest = 0;
        double fractionSolved = 0.0;
        foreach (var conf in GenerateConfigurations(instanceFactory, nReps, maxSizes)) {
          var prob = conf.Problem;
          var maxLen = conf.MaxSize;
          var rand = new Random(conf.RandSeed);

          var solver = new SequentialSearch(prob, maxLen, rand, 0,
            new GenericFunctionApproximationGrammarPolicy(prob, true));

          var problemName = prob.GetType().Name;
          double bestQ; int itersToBest;
          RunSolver(solver, problemName, string.Empty, 1.0, maxIterations, maxLen, out bestQ, out itersToBest);
          sumBestQ += bestQ;
          sumItersToBest += itersToBest;
          if (bestQ.IsAlmost(1.0)) fractionSolved += 1.0 / nReps;
        }
        // Assert.AreEqual(0.85, fractionSolved, 1E-6);
        // Assert.AreEqual(0.99438202247191, sumBestQ / nReps, 1E-6);
        // Assert.AreEqual(5461.7, sumItersToBest / (double)nReps, 1E-6);
      }
    }

    [TestMethod]
    [Timeout(1000 * 60 * 60 * 12)] // 12 hours
    // this configuration worked especially well in the experiments
    public void TestPoly10WithConstantOpt() {
      CultureInfo.DefaultThreadCurrentCulture = CultureInfo.InvariantCulture;

      var instanceFactories = new Func<int, ISymbolicExpressionTreeProblem>[]
      {
        (randSeed) => (ISymbolicExpressionTreeProblem)new SymbolicRegressionProblem(new Random(randSeed), "Poly-10", true ), 
      };

      var maxSizes = new int[] { 23 };  
      int nReps = 20;
      int maxIterations = 100000;
      foreach (var instanceFactory in instanceFactories) {
        var sumBestQ = 0.0;
        var sumItersToBest = 0;
        double fractionSolved = 0.0;
        foreach (var conf in GenerateConfigurations(instanceFactory, nReps, maxSizes)) {
          var prob = conf.Problem;
          var maxLen = conf.MaxSize;
          var rand = new Random(conf.RandSeed);

          var solver = new SequentialSearch(prob, maxLen, rand, 0,
            new GenericFunctionApproximationGrammarPolicy(prob, true));

          var problemName = prob.GetType().Name;
          double bestQ; int itersToBest;
          RunSolver(solver, problemName, string.Empty, 1.0, maxIterations, maxLen, out bestQ, out itersToBest);
          sumBestQ += bestQ;
          sumItersToBest += itersToBest;
          if (bestQ.IsAlmost(1.0)) fractionSolved += 1.0 / nReps;
        }
        // Assert.AreEqual(0.85, fractionSolved, 1E-6);
        // Assert.AreEqual(0.99438202247191, sumBestQ / nReps, 1E-6);
        // Assert.AreEqual(5461.7, sumItersToBest / (double)nReps, 1E-6);
      }
    }

    private IEnumerable<Configuration> GenerateConfigurations(Func<int, ISymbolicExpressionTreeProblem> problemFactory,
     int nReps,
     IEnumerable<int> maxSizes
     ) {
      var seedRand = new Random(randSeed);
      // the problem seed is the same for all configuratons 
      // this guarantees that we solve the _same_ problem each time
      // with different solvers and multiple repetitions
      var problemSeed = randSeed;
      for (int i = 0; i < nReps; i++) {
        // in each repetition use the same random seed for all solver configuratons
        // do nReps with different seeds for each configuration
        var solverSeed = seedRand.Next();
        foreach (var maxSize in maxSizes) {
          yield return new Configuration {
            MaxSize = maxSize,
            Problem = problemFactory(problemSeed),
            RandSeed = solverSeed
          };
        }
      }
    }




    private static void RunSolver(ISolver solver, string problemName, string policyName, double bestKnownQuality, int maxIters, int maxSize,
      out double bestQ, out int itersToBest) {
      int iterations = 0;
      var globalStatistics = new SentenceSetStatistics(bestKnownQuality);
      var solverName = solver.GetType().Name;
      double bestQuality = double.NegativeInfinity;
      int iterationsToBest = -1;
      solver.SolutionEvaluated += (sentence, quality) => {
        iterations++;
        globalStatistics.AddSentence(sentence, quality);
        if (quality > bestQuality) {
          bestQuality = quality;
          iterationsToBest = iterations;
        }
        if (iterations % 1000 == 0) {
          Console.WriteLine("\"{0,25}\" \"{1,25}\" {2} \"{3,25}\" {4}", solverName, policyName, maxSize, problemName, globalStatistics);
        }
      };


      solver.Run(maxIters);

      bestQ = bestQuality;
      itersToBest = iterationsToBest;
    }
  }
}