Changeset 16893

Timestamp:

05/04/19 08:41:14 (5 years ago)

Author:

gkronber

Message:

#2925: allow separate configuration of const opt steps for pre-tuning and the full ODE. Allow weighted combination of fitness using pretuning NMSE and the full ODE NMSE

Location:

branches/2925_AutoDiffForDynamicalModels/HeuristicLab.Problems.DynamicalSystemsModelling/3.3

Files:

• . (modified) (1 prop)
• Problem.cs (modified) (14 diffs)
• Solution.cs (modified) (5 diffs)
• sundials (modified) (1 prop)

branches/2925_AutoDiffForDynamicalModels/HeuristicLab.Problems.DynamicalSystemsModelling/3.3

@@ -1 +1 @@
 Plugin.cs
+sundials-x64

@@ -1 +1 @@
 Plugin.cs
+sundials-x64

branches/2925_AutoDiffForDynamicalModels/HeuristicLab.Problems.DynamicalSystemsModelling/3.3/Problem.cs

@@ -51 +51 @@
     private const string FunctionSetParameterName = "Function set";
     private const string MaximumLengthParameterName = "Size limit";
-    private const string MaximumParameterOptimizationIterationsParameterName = "Max. parameter optimization iterations";
+    private const string MaximumPretuningParameterOptimizationIterationsParameterName = "Max. pre-tuning parameter optimization iterations";
+    private const string MaximumOdeParameterOptimizationIterationsParameterName = "Max. ODE parameter optimization iterations";
     private const string NumberOfLatentVariablesParameterName = "Number of latent variables";
     private const string NumericIntegrationStepsParameterName = "Steps for numeric integration";
@@ -75 +76 @@
     }
 
-    public IFixedValueParameter<IntValue> MaximumParameterOptimizationIterationsParameter {
-      get { return (IFixedValueParameter<IntValue>)Parameters[MaximumParameterOptimizationIterationsParameterName]; }
+    public IFixedValueParameter<IntValue> MaximumPretuningParameterOptimizationIterationsParameter {
+      get { return (IFixedValueParameter<IntValue>)Parameters[MaximumPretuningParameterOptimizationIterationsParameterName]; }
+    }
+    public IFixedValueParameter<IntValue> MaximumOdeParameterOptimizationIterationsParameter {
+      get { return (IFixedValueParameter<IntValue>)Parameters[MaximumOdeParameterOptimizationIterationsParameterName]; }
     }
     public IFixedValueParameter<IntValue> NumberOfLatentVariablesParameter {
@@ -92 +96 @@
     public IConstrainedValueParameter<StringValue> OdeSolverParameter {
       get { return (IConstrainedValueParameter<StringValue>)Parameters[OdeSolverParameterName]; }
+    }
+    public IFixedValueParameter<DoubleValue> PretuningErrorWeight {
+      get { return (IFixedValueParameter<DoubleValue>)Parameters["Pretuning NMSE weight"]; }
+    }
+    public IFixedValueParameter<DoubleValue> OdeErrorWeight {
+      get { return (IFixedValueParameter<DoubleValue>)Parameters["ODE NMSE weight"]; }
     }
     #endregion
@@ -113 +123 @@
       get { return MaximumLengthParameter.Value.Value; }
     }
-    public int MaximumParameterOptimizationIterations {
-      get { return MaximumParameterOptimizationIterationsParameter.Value.Value; }
+    public int MaximumPretuningParameterOptimizationIterations {
+      get { return MaximumPretuningParameterOptimizationIterationsParameter.Value.Value; }
+    }
+    public int MaximumOdeParameterOptimizationIterations {
+      get { return MaximumOdeParameterOptimizationIterationsParameter.Value.Value; }
     }
     public int NumberOfLatentVariables {
@@ -155 +168 @@
         Parameters.Add(new FixedValueParameter<BoolValue>(OptimizeParametersForEpisodesParameterName, "Flag to select if parameters should be optimized globally or for each episode individually.", new BoolValue(false)));
       }
+      int iters = 100;
+      if (Parameters.ContainsKey("Max. parameter optimization iterations")) {
+        iters = ((IFixedValueParameter<IntValue>)Parameters["Max. parameter optimization iterations"]).Value.Value;
+      }
+      if (!Parameters.ContainsKey(MaximumPretuningParameterOptimizationIterationsParameterName)) {
+        Parameters.Add(new FixedValueParameter<IntValue>(MaximumPretuningParameterOptimizationIterationsParameterName, "The maximum number of iterations for optimization of parameters of individual equations for numerical derivatives (using L-BFGS). More iterations makes the algorithm slower, fewer iterations might prevent convergence in the optimization scheme. Default = 100", new IntValue(iters)));
+      }
+      if (!Parameters.ContainsKey(MaximumOdeParameterOptimizationIterationsParameterName)) {
+        Parameters.Add(new FixedValueParameter<IntValue>(MaximumOdeParameterOptimizationIterationsParameterName, "The maximum number of iterations for optimization of the full ODE parameters (using L-BFGS). More iterations makes the algorithm slower, fewer iterations might prevent convergence in the optimization scheme. Default = 100", new IntValue(iters)));
+      }
+
+      if (!Parameters.ContainsKey("Pretuning NMSE weight"))
+        Parameters.Add(new FixedValueParameter<DoubleValue>("Pretuning NMSE weight", "For fitness weighting", new DoubleValue(0.5)));
+      if (!Parameters.ContainsKey("ODE NMSE weight"))
+        Parameters.Add(new FixedValueParameter<DoubleValue>("ODE NMSE weight", "For fitness weighting", new DoubleValue(0.5)));
+
+
       RegisterEventHandlers();
     }
@@ -174 +204 @@
       Parameters.Add(new ValueParameter<ReadOnlyCheckedItemList<StringValue>>(FunctionSetParameterName, "The list of allowed functions", functions));
       Parameters.Add(new FixedValueParameter<IntValue>(MaximumLengthParameterName, "The maximally allowed length of each expression. Set to a small value (5 - 25). Default = 10", new IntValue(10)));
-      Parameters.Add(new FixedValueParameter<IntValue>(MaximumParameterOptimizationIterationsParameterName, "The maximum number of iterations for optimization of parameters (using L-BFGS). More iterations makes the algorithm slower, fewer iterations might prevent convergence in the optimization scheme. Default = 100", new IntValue(100)));
+      Parameters.Add(new FixedValueParameter<IntValue>(MaximumPretuningParameterOptimizationIterationsParameterName, "The maximum number of iterations for optimization of parameters of individual equations for numerical derivatives (using L-BFGS). More iterations makes the algorithm slower, fewer iterations might prevent convergence in the optimization scheme. Default = 100", new IntValue(100)));
+      Parameters.Add(new FixedValueParameter<IntValue>(MaximumOdeParameterOptimizationIterationsParameterName, "The maximum number of iterations for optimization of the full ODE parameters (using L-BFGS). More iterations makes the algorithm slower, fewer iterations might prevent convergence in the optimization scheme. Default = 100", new IntValue(100)));
       Parameters.Add(new FixedValueParameter<IntValue>(NumberOfLatentVariablesParameterName, "Latent variables (unobserved variables) allow us to produce expressions which are integrated up and can be used in other expressions. They are handled similarly to target variables in forward simulation / integration. The difference to target variables is that there are no data to which the calculated values of latent variables are compared. Set to a small value (0 .. 5) as necessary (default = 0)", new IntValue(0)));
       Parameters.Add(new FixedValueParameter<IntValue>(NumericIntegrationStepsParameterName, "Number of steps in the numeric integration that are taken from one row to the next (set to 1 to 100). More steps makes the algorithm slower, less steps worsens the accuracy of the numeric integration scheme.", new IntValue(10)));
       Parameters.Add(new ValueParameter<ItemList<IntRange>>(TrainingEpisodesParameterName, "A list of ranges that should be used for training, each range represents an independent episode. This overrides the TrainingSet parameter in ProblemData.", new ItemList<IntRange>()));
       Parameters.Add(new FixedValueParameter<BoolValue>(OptimizeParametersForEpisodesParameterName, "Flag to select if parameters should be optimized globally or for each episode individually.", new BoolValue(false)));
+      Parameters.Add(new FixedValueParameter<DoubleValue>("Pretuning NMSE weight", "For fitness weighting", new DoubleValue(0.5)));
+      Parameters.Add(new FixedValueParameter<DoubleValue>("ODE NMSE weight", "For fitness weighting", new DoubleValue(0.5)));
 
       var solversStr = new string[] { "HeuristicLab" /* , "CVODES" */};
@@ -207 +240 @@
         foreach (var episode in TrainingEpisodes) {
           // double[] optTheta;
-          double nmse = OptimizeForEpisodes(trees, problemData, targetVars, latentVariables, random, new[] { episode }, MaximumParameterOptimizationIterations, NumericIntegrationSteps, OdeSolver);
+          double nmse = OptimizeForEpisodes(trees, problemData, targetVars, latentVariables, random, new[] { episode }, MaximumPretuningParameterOptimizationIterations, NumericIntegrationSteps, OdeSolver, MaximumOdeParameterOptimizationIterations);
           // individual["OptTheta_" + eIdx] = new DoubleArray(optTheta); // write back optimized parameters so that we can use them in the Analysis method
           eIdx++;
@@ -216 +249 @@
       } else {
         // double[] optTheta;
-        double nmse = OptimizeForEpisodes(trees, problemData, targetVars, latentVariables, random, TrainingEpisodes, MaximumParameterOptimizationIterations, NumericIntegrationSteps, OdeSolver);
+        double nmse = OptimizeForEpisodes(trees, problemData, targetVars, latentVariables, random, TrainingEpisodes, MaximumPretuningParameterOptimizationIterations, NumericIntegrationSteps, OdeSolver, MaximumOdeParameterOptimizationIterations,
+          PretuningErrorWeight.Value.Value, OdeErrorWeight.Value.Value);
         // individual["OptTheta"] = new DoubleArray(optTheta); // write back optimized parameters so that we can use them in the Analysis method
         return nmse;
@@ -229 +263 @@
       IRandom random,
       IEnumerable<IntRange> episodes,
-      int maxParameterOptIterations,
+      int maxPretuningParameterOptIterations,
       int numericIntegrationSteps,
-      string odeSolver) {
+      string odeSolver,
+      int maxOdeParameterOptIterations,
+      double pretuningErrorWeight = 0.5,
+      double odeErrorWeight = 0.5
+      ) {
+
+
 
       // extract constants from trees (without trees for latent variables)
@@ -240 +280 @@
 
       // optimize parameters by fitting f(x,y) to calculated differences dy/dt(t)
-      double nmse = PreTuneParameters(trees, problemData, targetVars, latentVariables, random, episodes, maxParameterOptIterations,
+      double nmse = pretuningErrorWeight * PreTuneParameters(trees, problemData, targetVars, latentVariables, random, episodes, maxPretuningParameterOptIterations,
         initialTheta, out double[] pretunedParameters);
 
@@ -250 +290 @@
 
       // optimize parameters using integration of f(x,y) to calculate y(t)
-      nmse = OptimizeParameters(trees, problemData, targetVars, latentVariables, episodes, maxParameterOptIterations, pretunedParameters, numericIntegrationSteps, odeSolver,
+      nmse += odeErrorWeight * OptimizeParameters(trees, problemData, targetVars, latentVariables, episodes, maxOdeParameterOptIterations, pretunedParameters, numericIntegrationSteps, odeSolver,
         out double[] optTheta);
       // var optTheta = pretunedParameters;
@@ -285 +325 @@
       var latentTrees = trees.Take(latentVariables.Length).ToArray();
 
-      {
-        // first calculate values of latent variables by integration
+      // first calculate values of latent variables by integration
+      if(latentVariables.Length > 0) {
         var inputVariables = targetVars.Concat(latentTrees.SelectMany(t => t.IterateNodesPrefix().OfType<VariableTreeNode>().Select(n => n.VariableName))).Except(latentVariables).Distinct();
         var myState = new OptimizationData(latentTrees, targetVars, inputVariables.ToArray(), problemData, null, episodes.ToArray(), 10, latentVariables, "HeuristicLab");
@@ -334 +374 @@
         var paramCount = myState.nodeValueLookup.ParameterCount;
 
-        optTheta = new double[0];
-        if (initialTheta[treeIdx].Length > 0) {
+        optTheta = initialTheta[treeIdx];
+        if (initialTheta[treeIdx].Length > 0 && maxParameterOptIterations > -1) {
           try {
             alglib.minlmstate state;
@@ -398 +438 @@
       optTheta = initialTheta;
 
-      if (initialTheta.Length > 0) {
+      if (initialTheta.Length > 0 && maxParameterOptIterations > -1) {
         var lowerBounds = Enumerable.Repeat(-1000.0, initialTheta.Length).ToArray();
         var upperBounds = Enumerable.Repeat(1000.0, initialTheta.Length).ToArray();

branches/2925_AutoDiffForDynamicalModels/HeuristicLab.Problems.DynamicalSystemsModelling/3.3/Solution.cs

@@ -6 +6 @@
 using HeuristicLab.Data;
 using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
-using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
 using HeuristicLab.Problems.DataAnalysis;
 using HeuristicLab.Problems.DataAnalysis.Symbolic;
-using HeuristicLab.Random;
 using HEAL.Attic;
 
@@ -20 +18 @@
       get { return trees; }
     }
-    // [Storable]
-    // private double[] theta;
 
     [Storable]
@@ -59 +55 @@
       this.trees = new ISymbolicExpressionTree[original.trees.Length];
       for (int i = 0; i < trees.Length; i++) this.trees[i] = cloner.Clone(original.trees[i]);
-      // this.theta = new double[original.theta.Length];
-      // Array.Copy(original.theta, this.theta, this.theta.Length);
       this.problemData = cloner.Clone(original.problemData);
       this.targetVars = original.TargetVariables.ToArray();
@@ -89 +83 @@
     public IEnumerable<double[]> Predict(IntRange episode, int forecastHorizon) {
       var forecastEpisode = new IntRange(episode.Start, episode.End + forecastHorizon);
-      // 
-      // var random = new FastRandom(12345);
-      // snmse = Problem.OptimizeForEpisodes(trees, problemData, targetVars, latentVariables, random, new[] { forecastEpisode }, 100, numericIntegrationSteps, odeSolver);
 
       var inputVariables = trees.SelectMany(t => t.IterateNodesPrefix().OfType<VariableTreeNode>().Select(n => n.VariableName))
@@ -99 +90 @@
 
       var optimizationData = new Problem.OptimizationData(trees, targetVars, inputVariables.ToArray(), problemData, null, new[] { forecastEpisode }, numericIntegrationSteps, latentVariables, odeSolver);
-      // 
-      // 
-      // var theta = Problem.ExtractParametersFromTrees(trees);
-
-
       var fi = new double[forecastEpisode.Size * targetVars.Length];
       var jac = new double[forecastEpisode.Size * targetVars.Length, optimizationData.nodeValueLookup.ParameterCount];

@@ -1 +1 @@
 Plugin.cs
+sundials-x64