Example #1
0
 public FeatuerSet(
         SeedNodeSet seedNodeSet, FeatureExtractor extractor,
         ICollection<SelectedFragment> acceptingFragments, ICollection<SelectedFragment> rejectingFragments) {
     AcceptingFeatures =
             CreateAcceptingFeatures(seedNodeSet.AcceptedNodes, extractor,acceptingFragments)
                     .ToImmutableList();
     RejectingFeatures =
             CreateRejectingFeatures(seedNodeSet.RejectedNodes, extractor,rejectingFragments)
                     .ToImmutableList();
 }
Example #2
0
        public EncodingResult Encode(
                ICollection<string> codePaths, IEnumerable<CstNode> allCsts,
                LearningExperiment oracle, SeedNodeSet seedNodeSet = null) {
            var fileName = codePaths.Count > 0
                    ? string.Join(",", codePaths).GetHashCode() + "_" +
                      (codePaths.First() + "," + codePaths.Last() + ",").GetHashCode() + "_"
                      + codePaths.Count + ".encoded"
                    : null;
            var formatter = new BinaryFormatter();
            if (fileName != null && File.Exists(fileName)) {
                using (var fs = new FileStream(fileName, FileMode.Open, FileAccess.Read)) {
                    try {
                        var ret = ((EncodingResult)formatter.Deserialize(fs)).MakeImmutable();
                        Console.WriteLine("############### Warning ###############");
                        Console.WriteLine("Cache file of encoded result is used.");
                        Console.WriteLine("#######################################");
                        return ret;
                    } catch (Exception e) {
                        Console.Error.WriteLine(e);
                    }
                }
            }

            var allUppermostNodes = allCsts.SelectMany(
                    cst => LearningExperimentUtil.GetUppermostNodesByNames(cst, _selectedNodeNames));

            var result = new EncodingResult();
            if (seedNodeSet != null) {
                result.SeedAcceptedNodeCount = seedNodeSet.AcceptedNodes.Count;
                result.SeedNodeCount = result.SeedAcceptedNodeCount
                                       + seedNodeSet.RejectedNodes.Count;
                EncodeSeedNodes(
                        seedNodeSet.AcceptedNodes, result, result.IdealAcceptedVector2GroupPath,
                        result.SeedAcceptedVector2GroupPath, oracle);
                EncodeSeedNodes(
                        seedNodeSet.RejectedNodes, result, result.IdealRejectedVector2GroupPath,
                        result.SeedRejectedVector2GroupPath, oracle);
            }
            EncodeTargetNodes(allUppermostNodes, result, oracle);

            if (fileName != null) {
                using (var fs = new FileStream(fileName, FileMode.Create, FileAccess.Write)) {
                    formatter.Serialize(fs, result);
                }
            }
            return result.MakeImmutable();
        }
Example #3
0
        public LearningResult Learn(
                ICollection<string> seedPaths, ICollection<string> codePaths, string searchPattern,
                StreamWriter writer = null) {
            var allCsts = GenerateValidCsts(codePaths);
            var seedCsts = GenerateValidCsts(seedPaths).ToList();
            var seedNodes = seedCsts
                    .SelectMany(
                            cst => LearningExperimentUtil.GetUppermostNodesByNames(cst, OracleNames))
                    .Where(ProtectedIsAcceptedUsingOracle)
                    .ToList();

            var seedCst = seedCsts.First();
            var seedCode = seedCst.Code;
            var structuredCode = new StructuredCode(seedCode);

            var acceptingFragments = ConstructAcceptingFragments(structuredCode, seedCst, seedNodes);
            var rejectingFragments = ConstructRejectingFragments(structuredCode, seedCst);

            SeedNodeSet.Create(acceptingFragments, this);

            var preparingTime = Environment.TickCount;
            var extractor = CreateExtractor();
            var seedNodeSet = new SeedNodeSet(acceptingFragments.Select(f => f.Node), seedCsts, this);
            Console.WriteLine("#Accepted seed nodes: " + seedNodeSet.AcceptedNodes.Count
                              + " (" + acceptingFragments.Count + ")");
            Console.WriteLine("#Rejected seed nodes: " + seedNodeSet.RejectedNodes.Count
                              + " (" + rejectingFragments.Count + ")");

            var featureSet = new FeatuerSet(seedNodeSet, extractor, acceptingFragments, rejectingFragments);
            var groupPaths = seedNodeSet.SelectedNodeNames.Select(n => ">" + n + ">");
            var classifier = new Classifier(groupPaths, featureSet);
            Console.WriteLine(
                    "#Features: " + featureSet.AcceptingFeatureCount + ", "
                    + featureSet.RejectingFeatureCount);
            Console.WriteLine("Inner: " + extractor.IsInner);

            var featureEncoder = new FeatureEncoder(seedNodeSet.SelectedNodeNames, extractor,
                    featureSet);
            var encodingResult = featureEncoder.Encode(codePaths, allCsts, this, seedNodeSet);
            Console.WriteLine("#Unique Elements: " + encodingResult.VectorCount);
            if (encodingResult.IdealAcceptedVector2GroupPath.Keys.ToHashSet()
                    .Overlaps(encodingResult.IdealRejectedVector2GroupPath.Keys.ToHashSet())) {
                var others = encodingResult.IdealRejectedVector2GroupPath;
                var vector =
                        encodingResult.IdealAcceptedVector2GroupPath.Keys.First(others.ContainsKey);
                foreach (var featureString in featureEncoder.GetFeatureStringsByVector(vector)) {
                    Console.WriteLine(Experiment.Beautify(featureString));
                }
                throw new Exception("Master predicates can't classify elements!");
            }

            var groupCache = new GroupCache(encodingResult, classifier);
            var trainingSet = encodingResult.CreateTrainingVectorSet();
            classifier.Create(trainingSet, groupCache);
            Experiment.WriteFeatureStrings(Console.Out, classifier, featureEncoder);
            Console.WriteLine("Preparing time: " + (Environment.TickCount - preparingTime));

            var count = 0;
            var sumTime = Environment.TickCount;
            ClassificationResult classificationResult;
            while (true) {
                var time = Environment.TickCount;
                classificationResult = Classify(count, classifier, groupCache, encodingResult,
                        trainingSet);
                if (classificationResult.SuspiciousNodes == null) {
                    break;
                }

                var additionalAcceptedSet = RevealSuspiciousElements(
                        encodingResult.IdealAcceptedVector2GroupPath.Keys,
                        classificationResult.SuspiciousNodes, encodingResult, trainingSet);
                if (!classifier.Update(additionalAcceptedSet, trainingSet, groupCache)) {
                    count++;
                } else {
                    count = 0;
                }

                Console.WriteLine("Time: " + (Environment.TickCount - time));
            }
            classifier.MakeImmutable();
            Console.WriteLine();
            Console.WriteLine("Sum time: " + (Environment.TickCount - sumTime));
            var trainingVectorCount = trainingSet.Count;
            var idealVectorCount = encodingResult.IdealVectorSet.Count;
            Console.WriteLine("#Required vectors: " + trainingVectorCount + " / " + idealVectorCount);

            if (writer != null) {
                encodingResult.WriteResult(writer, trainingSet);
            }

            foreach (var groupPath in classifier.GroupPaths) {
                Console.WriteLine(groupPath);
            }

            classifier.Optimize(encodingResult.IdealRejectedVector2GroupPath.Keys, groupCache);

            return new LearningResult {
                ClassificationResult = classificationResult,
                Classifier = classifier,
                EncodingResult = encodingResult,
                FeatureEncoder = featureEncoder,
            };
        }