source: branches/Templates/RegressionProblemInstances/RegressionProblemInstance/RegressionProblemInstanceProvider.cs @ 18040

using System;
using System.Collections.Generic;
using HeuristicLab.Problems.DataAnalysis;
using HeuristicLab.Problems.Instances;
using HeuristicLab.Random;

namespace RegressionProblemInstances {
  public class RegressionProblemInstanceProvider : ProblemInstanceProvider<IRegressionProblemData> {
    public override IEnumerable<IDataDescriptor> GetDataDescriptors() {
      // if data is generated randomly then it is recommended to use a fixed random seed for reproducability
      // Here we even use a random number generator to produce the seeds for the instance specific random number generators
      var seedRandom = new MersenneTwister(1234);
      return new[]
      {
        new RegressionProblemInstanceDescriptor(
          name: "Poly-10", desc: "The Poly-10 problem.",
          rand: new NormalDistributedRandom(new MersenneTwister((uint) seedRandom.Next()), 0, 10), // x ~ N(0,10)
          dim: 10, numRows: 500,
          func: (double[] x) => x[0]*x[1] + x[2]*x[3] + x[4]*x[5] + x[0]*x[6]*x[8] + x[2]*x[5]*x[9]
          ),
        new RegressionProblemInstanceDescriptor(
          name: "sqrt(x+1)-sqrt(x)", desc: "Example from Herbie website http://herbie.uwplse.org/",
          rand: new UniformDistributedRandom(new MersenneTwister((uint) seedRandom.Next()), 0, 1E6),  // x ~ U(0, 1e6)
          dim: 1, numRows: 5000,
          func: (double[] x) => Math.Sqrt(x[0]+1) - Math.Sqrt(x[0])
          ),
        new RegressionProblemInstanceDescriptor(
          name: "Quadratic formula (-b/2a - sqrt(b^2 - 4ac)/2a", desc: "Example from Herbie paper http://herbie.uwplse.org/pldi15-paper.pdf",
          rand: new UniformDistributedRandom(new MersenneTwister((uint) seedRandom.Next()), -1E6, 1E6),  // x ~ U(-1e6, 1e6)
          dim: 3, numRows: 5000,
          func: (double[] x) => (-x[1] - Math.Sqrt(-x[1]*x[1]  - 4.0*x[0]*x[2])) / (2*x[0])
          )
      };
    }

    public override IRegressionProblemData LoadData(IDataDescriptor descriptor) {
      var desc = descriptor as IRegressionProblemInstanceDescriptor;
      if (desc == null) {
        throw new ArgumentException();
      } else {
        return desc.GenerateData();
      }
    }

    public override bool CanExportData { get { return false; } }
    public override bool CanImportData { get { return false; } }

    public override string Name { get { return "My problem instances"; } }
    public override string Description { get { return "A set of regression problem instance "; } }
    public override Uri WebLink { get { return new Uri("https://dev.heuristiclab.com"); } } // if there is a web uri with more information on the problem instances
    public override string ReferencePublication { get { return string.Empty; } } // if there is an academic publication for these problem instances
  }
}