Hill climbing for realistic graphs with consistency feature

5 files changed, 121 insertions, 7 deletions

diff --git a/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/app/Main.xtend b/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/app/Main.xtend
index cf871ead..1745bc35 100644
--- a/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/app/Main.xtend
+++ b/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/app/Main.xtend
@@ -35,7 +35,7 @@ class Main {
                 
                 //human input has different package declaration
 //              reader = new GraphReader(Yakindumm2PackageImpl.eINSTANCE);
-                val human = new RWInformation("inputs/Random/", "outputs/", 1);
+                val human = new RWInformation("inputs/Fake_Random_Random/", "outputs/", 1);
                 calculateAllModels(human.inputFolder, human.outputFolder,human.numRuns, reader);
                 
                 
diff --git a/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/app/PartialInterpretationMetric.xtend b/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/app/PartialInterpretationMetric.xtend
index cdd06027..71fa5fed 100644
--- a/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/app/PartialInterpretationMetric.xtend
+++ b/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/app/PartialInterpretationMetric.xtend
@@ -1,5 +1,6 @@
 package ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator.app
 
+import ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator.distance.JSDistance
 import ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator.distance.KSDistance
 import ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator.graph.PartialInterpretationGraph
 import ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator.io.CsvFileWriter
@@ -17,13 +18,14 @@ import org.eclipse.emf.ecore.util.EcoreUtil
 import org.eclipse.viatra.dse.api.Solution
 
 class PartialInterpretationMetric {
-        var static state = 0;
         var static KSDistance ks;
+        var static JSDistance js;
         
         def static void initPaths(){
                 new File("debug/metric/").mkdir();
                 new File("debug/metric/trajectories/").mkdir();
                 ks = new KSDistance(Domain.Yakinduum);
+                js = new JSDistance(Domain.Yakinduum);
         }
         
         def static MetricDistanceGroup calculateMetricDistance(PartialInterpretation partial){
@@ -35,6 +37,22 @@ class PartialInterpretationMetric {
                 val metricCalculator = new PartialInterpretationGraph(partial, metrics, null);          
                 var metricSamples = metricCalculator.evaluateAllMetricsToSamples();
                 
+                var mpc = js.mpcDistance(metricSamples.mpcSamples);
+                var na = js.naDistance(metricSamples.naSamples);
+                var outDegree = js.outDegreeDistance(metricSamples.outDegreeSamples);
+                
+                return new MetricDistanceGroup(mpc, na, outDegree);
+        }
+        
+        def static MetricDistanceGroup calculateMetricDistanceKS(PartialInterpretation partial){
+                val metrics = new ArrayList<Metric>();
+                metrics.add(new OutDegreeMetric());
+                metrics.add(new NodeActivityMetric());
+                metrics.add(new MultiplexParticipationCoefficientMetric());
+                
+                val metricCalculator = new PartialInterpretationGraph(partial, metrics, null);          
+                var metricSamples = metricCalculator.evaluateAllMetricsToSamples();
+                
                 var mpc = ks.mpcDistance(metricSamples.mpcSamples);
                 var na = ks.naDistance(metricSamples.naSamples);
                 var outDegree = ks.outDegreeDistance(metricSamples.outDegreeSamples);
@@ -52,7 +70,6 @@ class PartialInterpretationMetric {
                 //make dir since the folder can be none existing
                 new File(path).mkdir();
                 val filename = path + "/state_"+currentStateId+"-"+counter+".csv";
-                state++;
                 val metricCalculator = new PartialInterpretationGraph(partial, metrics, currentStateId);
                 
                 CsvFileWriter.write(metricCalculator.evaluateAllMetrics(), filename);
@@ -60,20 +77,22 @@ class PartialInterpretationMetric {
         
         def static void outputTrajectories(PartialInterpretation empty, List<Solution>  solutions){     
                 for(solution : solutions){
+                        
                         //need to copy the empty solution because the transition directly worked on the graph
                         val emptySolutionCopy = EcoreUtil.copy(empty)
                         val trajectory = solution.shortestTrajectory;
-                        trajectory.modelWithEditingDomain = emptySolutionCopy
+                        trajectory.model = emptySolutionCopy
                         
                         // state codes that will record the trajectory
                         val stateCodes = newArrayList()
-                        
                         var counter = 0
+
                         //transform and record the state codes for each state
                         while(trajectory.doNextTransformation){
                                 //println(trajectory.stateCoder.createStateCode)
                                 val stateId = trajectory.stateCoder.createStateCode.toString
                                 val interpretation = trajectory.getModel();
+                                println(stateId)
                                 //calculate metrics of current state
                                 calculateMetric(interpretation as PartialInterpretation, "debug/metric/output", stateId, counter)
                                 stateCodes.add(stateId)
diff --git a/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/distance/JSDistance.xtend b/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/distance/JSDistance.xtend
new file mode 100644
index 00000000..ced9eadb
--- /dev/null
+++ b/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/distance/JSDistance.xtend
@@ -0,0 +1,95 @@
+package ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator.distance
+
+import ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator.app.Domain
+import ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator.io.RepMetricsReader
+import com.google.common.collect.Sets
+import java.text.DecimalFormat
+import java.util.HashMap
+import java.util.List
+
+class JSDistance {
+        var HashMap<String, Double> mpcPMF;
+        var HashMap<String, Double> naPMF;
+        var HashMap<String, Double> outDegreePMF;
+        var DecimalFormat formatter;
+        
+        new(Domain d){
+                var metrics = RepMetricsReader.read(d);
+                var mpcSamples = metrics.mpcSamples;
+                var naSamples = metrics.naSamples.stream.mapToDouble([it]).toArray();
+                var outDegreeSamples = metrics.outDegreeSamples.stream.mapToDouble([it]).toArray();
+                
+                //needs to format the number to string avoid precision issue
+                formatter = new DecimalFormat("#0.00000");  
+                
+                mpcPMF = pmfFromSamples(mpcSamples, formatter);
+                naPMF = pmfFromSamples(naSamples, formatter);
+                outDegreePMF = pmfFromSamples(outDegreeSamples, formatter);     
+        }
+        
+        def private pmfFromSamples(double[] samples, DecimalFormat formatter){
+                var length = samples.length;
+                var pmfMap = new HashMap<String, Double>();
+                
+                for(sample : samples){
+                        pmfMap.put(formatter.format(sample), pmfMap.getOrDefault(formatter.format(sample), 0.0) + 1.0 / length);
+                }
+                
+                return pmfMap;
+        }
+        
+        def private combinePMF(HashMap<String, Double> pmf1, HashMap<String, Double> pmf2){
+                var pmfMap = new HashMap<String, Double>();
+                
+                var union = Sets.union(pmf1.keySet(), pmf2.keySet());
+                
+                for(key : union){
+                        // corresponding to M in JS distance
+                        var value = 1.0/2 * (pmf1.getOrDefault(key, 0.0) + pmf2.getOrDefault(key, 0.0));
+                        pmfMap.put(key, value);
+                }
+                return pmfMap;
+        }
+        
+        def private jsDivergence(HashMap<String, Double> p, HashMap<String, Double> q){
+                val m = combinePMF(q, p);
+                var distance = 1.0/2 * klDivergence(p, m) + 1.0/2 * klDivergence(q, m);
+                return distance;
+        }
+        
+        def klDivergence(HashMap<String, Double> p, HashMap<String, Double> q){
+                var distance = 0.0;
+                for(key : q.keySet()){
+                        //need to convert log e to log 2
+                        if(p.containsKey(key)){
+                                distance -= p.get(key) * Math.log(q.get(key) / p.get(key)) / Math.log(2);
+                        }
+                }
+                return distance;
+        }
+        
+        def double mpcDistance(List<Double> samples){
+                // map list to array
+                var map = pmfFromSamples(samples.stream().mapToDouble([it]).toArray(), formatter);
+                //if the size of array is smaller than 2, ks distance cannot be performed, simply return 1
+                if(map.size < 2) return 1;
+                return jsDivergence(map, mpcPMF);
+        }
+        
+        def double naDistance(List<Double> samples){
+                // map list to array
+                var map = pmfFromSamples(samples.stream().mapToDouble([it]).toArray(), formatter);
+                
+                //if the size of array is smaller than 2, ks distance cannot be performed, simply return 1
+                if(map.size < 2) return 1;
+                return jsDivergence(map, naPMF);
+        }
+        
+        def double outDegreeDistance(List<Double> samples){
+                // map list to array
+                var map = pmfFromSamples(samples.stream().mapToDouble([it]).toArray(), formatter);
+                //if the size of array is smaller than 2, ks distance cannot be performed, simply return 1
+                if(map.size < 2) return 1;
+                return jsDivergence(map, outDegreePMF);
+        }
+}
\ No newline at end of file
diff --git a/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/distance/KSDistance.xtend b/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/distance/KSDistance.xtend
index 1fb21529..58e0a8a3 100644
--- a/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/distance/KSDistance.xtend
+++ b/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/distance/KSDistance.xtend
@@ -1,9 +1,9 @@
 package ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator.distance
 
 import ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator.app.Domain
+import ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator.io.RepMetricsReader
 import java.util.List
 import org.apache.commons.math3.stat.inference.KolmogorovSmirnovTest
-import ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator.io.RepMetricsReader
 
 class KSDistance {
         var static ksTester = new KolmogorovSmirnovTest();
diff --git a/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/io/CsvFileWriter.xtend b/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/io/CsvFileWriter.xtend
index bed356e9..01e3940b 100644
--- a/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/io/CsvFileWriter.xtend
+++ b/Metrics/Metrics-Calculation/ca.mcgill.ecse.dslreasoner.realistic.metrics.calculator/src/ca/mcgill/ecse/dslreasoner/realistic/metrics/calculator/io/CsvFileWriter.xtend
@@ -21,7 +21,7 @@ class CsvFileWriter {
                                         output.append(datarow.get(i) + ',');
                                 }
                                 
-                                if(datarow.size > 1){
+                                if(datarow.size >= 1){
                                         output.append(datarow.get(datarow.size() - 1));
                                         output.append('\n');
                                 }