source: branches/2679_HeuristicLab.GoalSeekingProblem/HeuristicLab.GoalSeekingProblem/3.4/SingleObjectiveGoalSeekingProblem.cs @ 17869

#region License Information
/* HeuristicLab
 * Copyright (C) 2002-2016 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 HEAL.Attic;
using HeuristicLab.Collections;
using HeuristicLab.Common;
using HeuristicLab.Core;
using HeuristicLab.Data;
using HeuristicLab.Encodings.RealVectorEncoding;
using HeuristicLab.Optimization;
using HeuristicLab.Parameters;
using HeuristicLab.Problems.DataAnalysis;
using System;
using System.Collections.Generic;
using System.Linq;

namespace HeuristicLab.GoalSeeking {
  [Item("Goal seeking problem (single-objective)", "Represents a single objective optimization problem which uses configurable regression solutions to evaluate targets from a given dataset.")]
  [Creatable("Problems")]
  [StorableType("DD67A460-1A32-414A-AB18-E0C773B46689")]
  public sealed class SingleObjectiveGoalSeekingProblem : SingleObjectiveBasicProblem<RealVectorEncoding>, IGoalSeekingProblem {
    #region parameter names
    private const string InputsParameterName = "Inputs";
    private const string GoalsParameterName = "Goals";
    private const string ModelsParameterName = "Models";
    #endregion

    #region parameters
    public IValueParameter<CheckedItemList<InputParameter>> InputsParameter {
      get { return (IValueParameter<CheckedItemList<InputParameter>>)Parameters[InputsParameterName]; }
    }
    public IValueParameter<CheckedItemList<GoalParameter>> GoalsParameter {
      get { return (IValueParameter<CheckedItemList<GoalParameter>>)Parameters[GoalsParameterName]; }
    }
    public IFixedValueParameter<ItemList<IRegressionModel>> ModelsParameter {
      get { return (IFixedValueParameter<ItemList<IRegressionModel>>)Parameters[ModelsParameterName]; }
    }
    #endregion

    #region IGoalSeekingProblem implementation
    public IEnumerable<IRegressionModel> Models {
      get { return ModelsParameter.Value; }
    }

    public IEnumerable<GoalParameter> Goals {
      get { return GoalsParameter.Value; }
    }

    public IEnumerable<InputParameter> Inputs {
      get { return InputsParameter.Value; }
    }

    public void AddModel(IRegressionModel model) {
      var models = ModelsParameter.Value;
      models.Add(model);
      GoalSeekingUtil.RaiseEvent(this, ModelsChanged);
    }

    public void RemoveModel(IRegressionModel model) {
      var models = ModelsParameter.Value;
      models.Remove(model);
      GoalSeekingUtil.RaiseEvent(this, ModelsChanged);
    }

    public void Configure(IRegressionProblemData problemData, int row) {
      GoalSeekingUtil.Configure(Goals, Inputs, problemData, row);
    }

    public IEnumerable<double> GetEstimatedGoalValues(IEnumerable<double> parameterValues, bool round = false) {
      var ds = (ModifiableDataset)dataset.Clone();
      foreach (var parameter in ActiveInputs.Zip(parameterValues, (p, v) => new { Name = p.Name, Value = v })) {
        ds.SetVariableValue(parameter.Value, parameter.Name, 0);
      }
      var rows = new[] { 0 }; // actually just one row
      return round ? ActiveGoals.Select(t => RoundToNearestStepMultiple(GetModels(t.Name).Average(m => m.GetEstimatedValues(ds, rows).Single()), t.Step))
                   : ActiveGoals.Select(t => GetModels(t.Name).Average(m => m.GetEstimatedValues(ds, rows).Single()));
    }

    public event EventHandler ModelsChanged;
    public event EventHandler TargetsChanged;
    public event EventHandler ParametersChanged;
    #endregion

    private IEnumerable<GoalParameter> ActiveGoals {
      get { return Goals.Where(x => x.Active); }
    }
    private IEnumerable<InputParameter> ActiveInputs {
      get { return Inputs.Where(x => x.Active); }
    }

    [Storable]
    private ModifiableDataset dataset; // modifiable dataset

    public override bool Maximization {
      get { return false; }
    }

    #region constructors
    [StorableConstructor]
    private SingleObjectiveGoalSeekingProblem(StorableConstructorFlag _) : base(_) { }

    private SingleObjectiveGoalSeekingProblem(SingleObjectiveGoalSeekingProblem original, Cloner cloner) : base(original, cloner) {
      this.dataset = cloner.Clone(original.dataset);
      RegisterEvents();
    }

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

    [StorableHook(HookType.AfterDeserialization)]
    private void AfterDeserialization() {
      RegisterEvents();
    }

    public SingleObjectiveGoalSeekingProblem() {
      dataset = new ModifiableDataset();
      Parameters.Add(new ValueParameter<CheckedItemList<InputParameter>>(InputsParameterName));
      Parameters.Add(new ValueParameter<CheckedItemList<GoalParameter>>(GoalsParameterName));
      Parameters.Add(new FixedValueParameter<ItemList<IRegressionModel>>(ModelsParameterName, new ItemList<IRegressionModel>()));
      EncodingParameter.Hidden = true;
      EvaluatorParameter.Hidden = true;
      SolutionCreatorParameter.Hidden = true;
      RegisterEvents();
    }
    #endregion

    public override double Evaluate(Individual individual, IRandom random) {
      var vector = individual.RealVector();
      vector.ElementNames = ActiveInputs.Select(x => x.Name);
      int i = 0;
      // round vector according to parameter step sizes
      foreach (var parameter in ActiveInputs) {
        vector[i] = RoundToNearestStepMultiple(vector[i], parameter.Step);
        ++i;
      }
      var estimatedValues = GetEstimatedGoalValues(vector, round: true);
      return ActiveGoals.Zip(estimatedValues, (t, v) => new { Target = t, EstimatedValue = v })
                        .Average(x => x.Target.Weight * Math.Pow(x.EstimatedValue - x.Target.Goal, 2) / x.Target.Variance);
    }

    public override void Analyze(Individual[] individuals, double[] qualities, ResultCollection results, IRandom random) {
      var zipped = individuals.Zip(qualities, (ind, qual) => new { Individual = ind, Quality = qual }).OrderBy(x => x.Quality);
      var best = Maximization ? zipped.Last() : zipped.First();
      var realVector = best.Individual.RealVector();
      const string resultName = "Best Solution";

      var columnNames = new List<string>();
      foreach (var goal in ActiveGoals) {
        columnNames.Add(goal.Name);
        columnNames.Add(goal.Name + " (estimated)");
      }
      foreach (var input in ActiveInputs) {
        columnNames.Add(input.Name);
        columnNames.Add(input.Name + " (estimated)");
        columnNames.Add(input.Name + " (deviation)");
      }

      var m = new DoubleMatrix(1, columnNames.Count) { ColumnNames = columnNames };
      int i = 0;

      var goals = ActiveGoals.Zip(GetEstimatedGoalValues(realVector, round: true),
        (goal, value) => new { TargetValue = goal.Goal, EstimatedValue = value });

      foreach (var goal in goals) {
        m[0, i] = goal.TargetValue;
        m[0, i + 1] = goal.EstimatedValue;
        i += 2;
      }

      var inputs = ActiveInputs.Zip(realVector,
        (input, value) => new { ActualValue = input.Value, EstimatedValue = value });

      foreach (var input in inputs) {
        m[0, i] = input.ActualValue;
        m[0, i + 1] = input.EstimatedValue;
        m[0, i + 2] = m[0, i] - m[0, i + 1];
        i += 3;
      }

      if (!results.ContainsKey(resultName)) {
        results.Add(new Result(resultName, m));
      } else {
        results[resultName].Value = m;
      }

      base.Analyze(individuals, qualities, results, random);
    }

    #region event handlers
    private void RegisterEvents() {
      ModelsParameter.Value.ItemsAdded += ModelCollection_ItemsChanged;
      ModelsParameter.Value.ItemsRemoved += ModelCollection_ItemsChanged;
      GoalsParameter.Value.CheckedItemsChanged += GoalSeekingUtil.Goals_CheckedItemsChanged;
      InputsParameter.Value.CheckedItemsChanged += GoalSeekingUtil.Inputs_CheckedItemsChanged;

      foreach (var input in Inputs)
        input.Changed += InputParameterChanged;

      foreach (var goal in Goals)
        goal.Changed += GoalParameterChanged;
    }

    private void ModelCollection_ItemsChanged(object sender, CollectionItemsChangedEventArgs<IndexedItem<IRegressionModel>> e) {
      if (e.Items == null || !e.Items.Any()) return;
      GoalSeekingUtil.UpdateInputs(InputsParameter.Value, Models, InputParameterChanged);
      GoalSeekingUtil.UpdateEncoding(Encoding, ActiveInputs);
      dataset = Inputs.Any() ? new ModifiableDataset(Inputs.Select(x => x.Name), Inputs.Select(x => new List<double> { x.Value })) : new ModifiableDataset();
      GoalSeekingUtil.UpdateTargets(GoalsParameter.Value, Models, GoalParameterChanged);
      GoalSeekingUtil.RaiseEvent(this, ModelsChanged);
    }

    private void InputParameterChanged(object sender, EventArgs args) {
      var inputParameter = (InputParameter)sender;
      var inputs = InputsParameter.Value;
      if (inputs.ItemChecked(inputParameter) != inputParameter.Active)
        inputs.SetItemCheckedState(inputParameter, inputParameter.Active);
      GoalSeekingUtil.UpdateEncoding(Encoding, ActiveInputs);
    }

    private void GoalParameterChanged(object sender, EventArgs args) {
      var goalParameter = (GoalParameter)sender;
      var goals = GoalsParameter.Value;
      if (goals.ItemChecked(goalParameter) != goalParameter.Active)
        goals.SetItemCheckedState(goalParameter, goalParameter.Active);
    }
    #endregion

    #region helper methods
    // method which throws an exception that can be caught in the event handler if the check fails
    private void CheckIfDatasetContainsTarget(string target) {
      if (dataset.DoubleVariables.All(x => x != target))
        throw new ArgumentException(string.Format("Model target \"{0}\" does not exist in the dataset.", target));
    }

    private IEnumerable<IRegressionModel> GetModels(string target) {
      return Models.Where(x => x.TargetVariable == target);
    }

    private static double RoundToNearestStepMultiple(double value, double step) {
      return step * (long)Math.Round(value / step);
    }
    #endregion
  }
}