public SimulatedAnnealing(Learner l)
{
SimulatedAnnealingRunningParameters pam;
using (var file = File.OpenText(@"SimulatedAnnealingParameters.json"))
{
var serializer = new JsonSerializer();
pam =
(SimulatedAnnealingRunningParameters)
serializer.Deserialize(file, typeof(SimulatedAnnealingRunningParameters));
}
learner = l;
threshold = pam.ThresholdTemperature;
coolingFactor = pam.CoolingFactor;
currentIteration = bestIteration = 1;
reportEveryNIterations = pam.ReportEveryNIteration;
currentHypothesis = bestHypothesis = learner.GetInitialGrammar();
currentHypothesis.GenerateDerivedRulesFromSchema();
currentEnergy = bestEnergy = learner.Energy(currentHypothesis);
currentHypothesis.GenerateInitialRulesFromDerivedRules();
currentTemp = currentEnergy.TotalEnergy*pam.InitialTemperatureTimesInitialEnegrgy;
using (var sw = File.AppendText("SessionReport.txt"))
{
sw.WriteLine(string.Format("cooling factor: {0}, initial energy: {1}, initial temperature: {2}",
coolingFactor, currentEnergy, currentTemp));
}
}
public static EnergyData TotalLength(Grammar grammar, Tuple<Node, int>[] parseTreesWithCounts)
{
var e = new EnergyData();
//var rules = grammar.ruleNumberDictionary.Values;
e.GrammarEnergy = GetGrammarLength(grammar);
e.DataEnergy = GetDataLength(grammar, parseTreesWithCounts);
return e;
}
public static EnergyData TotalLength(Grammar grammar, Tuple <Node, int>[] parseTreesWithCounts)
{
var e = new EnergyData();
//var rules = grammar.ruleNumberDictionary.Values;
e.GrammarEnergy = GetGrammarLength(grammar);
e.DataEnergy = GetDataLength(grammar, parseTreesWithCounts);
return(e);
}
public Tuple<EnergyData, Grammar> Run()
{
var rand = new Random();
while (currentTemp > threshold)
{
try
{
var newHypothesis = learner.GetNeighbor(currentHypothesis);
EnergyData newEnergy = null;
if (newHypothesis != null)
{
newHypothesis.GenerateDerivedRulesFromSchema();
newEnergy = learner.Energy(newHypothesis);
newHypothesis.GenerateInitialRulesFromDerivedRules();
}
if (newEnergy != null)
{
if (newEnergy < bestEnergy)
{
bestEnergy = newEnergy;
bestHypothesis = newHypothesis;
bestIteration = currentIteration;
}
var prob = P(currentEnergy, newEnergy, currentTemp);
if (rand.NextDouble() < prob)
{
// moved to new hypothesis
currentHypothesis = newHypothesis;
currentEnergy = newEnergy;
}
}
currentIteration++;
if (currentIteration%reportEveryNIterations == 0)
Console.WriteLine("Iteration {0}", currentIteration);
currentTemp *= coolingFactor;
}
catch (Exception e)
{
Console.WriteLine(e.ToString());
}
}
var actualRuleDistributions = learner.CollectUsages(bestHypothesis);
var s = string.Format("{0}. ({1}) \nBest Hypothesis: \n{2}Best so far: #{3} with energy: {4}\n",
currentIteration, currentTemp, bestHypothesis.GrammarWithRuleUsages(actualRuleDistributions),
bestIteration, bestEnergy);
Console.WriteLine(s);
using (var sw = File.AppendText("SessionReport.txt"))
{
sw.WriteLine(s);
}
return new Tuple<EnergyData, Grammar>(bestEnergy, bestHypothesis);
}
private double P(EnergyData currStateEnergy, EnergyData possibleStateEnergy, double temp)
{
var calcValue = Math.Exp((currStateEnergy.TotalEnergy - possibleStateEnergy.TotalEnergy)/temp);
return Math.Min(1.0, calcValue);
}