source: branches/3076_IA_evaluators_analyzers/HeuristicLab.Problems.DataAnalysis.Symbolic.Regression/3.4/MultiObjective/SymbolicRegressionMultiObjectiveMultiSoftConstraintEvaluator.cs @ 17743

#region License Information

/* HeuristicLab
 * Copyright (C) Heuristic and Evolutionary Algorithms Laboratory (HEAL)
 *
 * This file is part of HeuristicLab.
 *
 * HeuristicLab is free software: you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 *
 * HeuristicLab is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
 */

#endregion

using System;
using System.Collections.Generic;
using System.Linq;
using HEAL.Attic;
using HeuristicLab.Common;
using HeuristicLab.Core;
using HeuristicLab.Data;
using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
using HeuristicLab.Parameters;

namespace HeuristicLab.Problems.DataAnalysis.Symbolic.Regression.MultiObjective {
  [Item("Multi Soft Constraints Evaluator",
    "Calculates the Person R² and the constraints violations of a symbolic regression solution.")]
  [StorableType("8E9D76B7-ED9C-43E7-9898-01FBD3633880")]
  public class
    SymbolicRegressionMultiObjectiveMultiSoftConstraintEvaluator : SymbolicRegressionMultiObjectiveSplittingEvaluator {
    public const string DimensionsParameterName = "Dimensions";

    public IFixedValueParameter<IntValue> DimensionsParameter => (IFixedValueParameter<IntValue>)Parameters[DimensionsParameterName];


    #region Constructors
    public SymbolicRegressionMultiObjectiveMultiSoftConstraintEvaluator() {
      Parameters.Add(new FixedValueParameter<IntValue>(DimensionsParameterName, new IntValue(2)));
    }

    [StorableConstructor]
    protected SymbolicRegressionMultiObjectiveMultiSoftConstraintEvaluator(StorableConstructorFlag _) : base(_) { }

    protected SymbolicRegressionMultiObjectiveMultiSoftConstraintEvaluator(
      SymbolicRegressionMultiObjectiveMultiSoftConstraintEvaluator original, Cloner cloner) : base(original, cloner) { }

    #endregion

    [StorableHook(HookType.AfterDeserialization)]
    private void AfterDeserialization() {
      if(!Parameters.ContainsKey(DimensionsParameterName))
        Parameters.Add(new FixedValueParameter<IntValue>(DimensionsParameterName, new IntValue(2)));
    }

    public override IDeepCloneable Clone(Cloner cloner) {
      return new SymbolicRegressionMultiObjectiveMultiSoftConstraintEvaluator(this, cloner);
    }


    public override IOperation InstrumentedApply() {
      var rows                  = GenerateRowsToEvaluate();
      var solution              = SymbolicExpressionTreeParameter.ActualValue;
      var problemData           = ProblemDataParameter.ActualValue;
      var interpreter           = SymbolicDataAnalysisTreeInterpreterParameter.ActualValue;
      var estimationLimits      = EstimationLimitsParameter.ActualValue;
      var minIntervalWidth      = MinSplittingWidth;
      var maxIntervalSplitDepth = MaxSplittingDepth;
      var applyLinearScaling  = ApplyLinearScalingParameter.ActualValue.Value;

      if (applyLinearScaling) {
        //Check for interval arithmetic grammar
        //remove scaling nodes for linear scaling evaluation
        var                    rootNode  = new ProgramRootSymbol().CreateTreeNode();
        var                    startNode = new StartSymbol().CreateTreeNode();
        SymbolicExpressionTree newTree   = null;
        foreach (var node in solution.IterateNodesPrefix())
          if (node.Symbol.Name == "Scaling") {
            for (var i = 0; i < node.SubtreeCount; ++i) startNode.AddSubtree(node.GetSubtree(i));
            rootNode.AddSubtree(startNode);
            newTree = new SymbolicExpressionTree(rootNode);
            break;
          }

        //calculate alpha and beta for scaling
        var estimatedValues = interpreter.GetSymbolicExpressionTreeValues(newTree, problemData.Dataset, rows);
        var targetValues    = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, rows);
        OnlineLinearScalingParameterCalculator.Calculate(estimatedValues, targetValues, out var alpha, out var beta,
          out var errorState);
        //Set alpha and beta to the scaling nodes from ia grammar
        foreach (var node in solution.IterateNodesPrefix())
          if (node.Symbol.Name == "Offset") {
            node.RemoveSubtree(1);
            var alphaNode = new ConstantTreeNode(new Constant()) {Value = alpha};
            node.AddSubtree(alphaNode);
          } else if (node.Symbol.Name == "Scaling") {
            node.RemoveSubtree(1);
            var betaNode = new ConstantTreeNode(new Constant()) {Value = beta};
            node.AddSubtree(betaNode);
          }
      }

      if (UseConstantOptimization)
        SymbolicRegressionConstantOptimizationEvaluator.OptimizeConstants(interpreter, solution, problemData, rows,
          false,
          ConstantOptimizationIterations,
          ConstantOptimizationUpdateVariableWeights,
          estimationLimits.Lower,
          estimationLimits.Upper);

      var qualities = Calculate(interpreter, solution, estimationLimits.Lower, estimationLimits.Upper, problemData,
        rows, UseIntervalSplitting);
      QualitiesParameter.ActualValue = new DoubleArray(qualities);
      return base.InstrumentedApply();
    }

    public override double[] Evaluate(IExecutionContext context, ISymbolicExpressionTree tree,
                                      IRegressionProblemData problemData,
                                      IEnumerable<int> rows) {
      SymbolicDataAnalysisTreeInterpreterParameter.ExecutionContext = context;
      EstimationLimitsParameter.ExecutionContext                    = context;
      ApplyLinearScalingParameter.ExecutionContext                  = context;

      var quality = Calculate(SymbolicDataAnalysisTreeInterpreterParameter.ActualValue, tree,
        EstimationLimitsParameter.ActualValue.Lower, EstimationLimitsParameter.ActualValue.Upper,
        problemData, rows, UseIntervalSplitting);

      SymbolicDataAnalysisTreeInterpreterParameter.ExecutionContext = null;
      EstimationLimitsParameter.ExecutionContext                    = null;
      ApplyLinearScalingParameter.ExecutionContext                  = null;

      return quality;
    }


    public static double[] Calculate(ISymbolicDataAnalysisExpressionTreeInterpreter interpreter,
                                     ISymbolicExpressionTree solution, double lowerEstimationLimit,
                                     double upperEstimationLimit,
                                     IRegressionProblemData problemData, IEnumerable<int> rows, bool splitting) {
      OnlineCalculatorError errorState;
      var estimatedValues =
        interpreter.GetSymbolicExpressionTreeValues(solution, problemData.Dataset, rows);
      var targetValues = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, rows);

      double nmse;

      var boundedEstimatedValues = estimatedValues.LimitToRange(lowerEstimationLimit, upperEstimationLimit);
      nmse = OnlineNormalizedMeanSquaredErrorCalculator.Calculate(targetValues, boundedEstimatedValues, out errorState);
      if (errorState != OnlineCalculatorError.None) nmse = 1.0;

      if (nmse > 1)
        nmse = 1.0;

      var constraints    = problemData.IntervalConstraints.Constraints.Where(c => c.Enabled);
      var variableRanges = problemData.VariableRanges.GetReadonlyDictionary();

      var objectives          = new List<double> {nmse};
      var intervalInterpreter = new IntervalInterpreter(){UseIntervalSplitting = splitting};

      var constraintObjectives = new List<double>();
      foreach (var c in constraints) {
        var penalty = ConstraintExceeded(c, intervalInterpreter, variableRanges,
          solution);
        var maxP = 0.1;

        if (double.IsNaN(penalty) || double.IsInfinity(penalty) || penalty > maxP)
          penalty = maxP;

        constraintObjectives.Add(penalty);
      }

      objectives.AddRange(constraintObjectives);

      return objectives.ToArray();
    }

    public static double ConstraintExceeded(IntervalConstraint constraint, IntervalInterpreter intervalInterpreter,
                                            IReadOnlyDictionary<string, Interval> variableRanges,
                                            ISymbolicExpressionTree solution) {
      if (constraint.Variable != null && !variableRanges.ContainsKey(constraint.Variable))
        throw new ArgumentException(
          $"The given variable {constraint.Variable} in the constraint does not exists in the model.",
          nameof(IntervalConstraintsParser));
      Interval resultInterval;
      if (!constraint.IsDerivative) {
        resultInterval =
          intervalInterpreter.GetSymbolicExpressionTreeInterval(solution, variableRanges, 0);
      }
      else {
        var tree = solution;
        for (var i = 0; i < constraint.NumberOfDerivations; ++i)
          tree = DerivativeCalculator.Derive(tree, constraint.Variable);

        resultInterval =
          intervalInterpreter.GetSymbolicExpressionTreeInterval(tree, variableRanges, 0);
      }
      
      //Calculate soft-constraints for intervals
      if (constraint.Interval.Contains(resultInterval)) return 0;
      var pLower = 0.0;
      var pUpper = 0.0;
      if (constraint.Interval.Contains(resultInterval.LowerBound))
        pLower = 0;
      else
        pLower = constraint.Interval.LowerBound - resultInterval.LowerBound;

      if (constraint.Interval.Contains(resultInterval.UpperBound))
        pUpper = 0;
      else
        pUpper = resultInterval.UpperBound - constraint.Interval.UpperBound;
      var penalty = Math.Abs(pLower) + Math.Abs(pUpper);

      return penalty;
    }

    /*
     * First objective is to maximize the Pearson R² value
     * All following objectives have to be minimized ==> Constraints
     */
    public override IEnumerable<bool> Maximization {
      get {
        var objectives = new List<bool> {false};          //First NMSE ==> min
        objectives.AddRange(Enumerable.Repeat(false, DimensionsParameter.Value.Value)); //Constraints ==> min
        
        return objectives;
      }
    }
  }
}