/// <summary>
/// Run the recognizer on a sketch.
///
/// This method is multithreaded.
///
/// Precondition: the sketch has been grouped into shapes.
///
/// Postcondition: shape.Classification has been filled in with this recognizer's best guess
/// for every shape in the sketch.
/// </summary>
/// <param name="featureSketch">the sketch to work on</param>
public override void process(Featurefy.FeatureSketch featureSketch)
{
List <Thread> threads = new List <Thread>(_numThreads);
int shapesPerWorker = featureSketch.NumShapes / _numThreads;
int finalEnd = 0;
// fork off (_numThreads - 1) threads, each with shapesPerWorker tasks to perform
for (int i = 0; i < _numThreads - 1; i++)
{
int start = i * shapesPerWorker;
int end = start + shapesPerWorker;
finalEnd = end;
Worker worker = new Worker(_innerRecognizer, start, end, featureSketch);
Thread t = new Thread(worker.run);
t.Name = "recognition worker thread " + i;
threads.Add(t);
t.Start();
}
// do the rest of the work on this thread
Worker myWorker = new Worker(_innerRecognizer, finalEnd, featureSketch.NumShapes, featureSketch);
myWorker.run();
// join the other threads to make sure they finished
foreach (Thread t in threads)
{
t.Join();
}
}
/// <summary>
/// Look at any out-of-context shapes in the sketch and refine them by
/// removing a substroke. (We may want to try removing more strokes later).
///
/// We assume that the substroke should be a wire, and we remove it
/// from the shape. Then we regroup the shape without the substroke,
/// re-recognize it, and check to see if it makes sense. If it does,
/// keep it. If not, try again with a different substroke.
/// </summary>
public void StrokeShedRefine(Featurefy.FeatureSketch featureSketch)
{
foreach (Sketch.Shape shape in featureSketch.Sketch.Shapes)
{
bool valid = _domain.IsProperlyConnected(shape);
// If the shape wansn't specified by the user, doesn't fit
// its context, and is a gate
if ((!shape.AlreadyGrouped) &&
(!valid || (shape.Probability < _thresholdProbability)) &&
(shape.Classification == LogicDomain.GATE_CLASS))
{
float bestProb = 0F;
Sketch.Substroke bestSubstroke = null;
foreach (Sketch.Substroke substroke in shape.Substrokes)
{
// See how helpful it is to let the shape
// shed this stroke
float shedProb = strokeShedHelpfulness(shape, substroke, featureSketch);
if (shedProb > bestProb)
{
bestProb = shedProb;
bestSubstroke = substroke;
}
}
// If hypotheticaly shedding one of the substrokes
// enhanced the recognition of the shape, do it for real
//if (bestProb > shape.Probability) // NOT SURE IF WE WANT TO KEEP THIS
shedStroke(shape, bestSubstroke, featureSketch);
}
}
}
/// <summary>
/// Recognizes a given shape and updates the shape accordingly.
/// </summary>
/// <param name="shape">The shape to recogize</param>
/// <param name="featureSketch">A featuresketch</param>
public override void recognize(Sketch.Shape shape, Featurefy.FeatureSketch featureSketch)
{
// Make sure we do nothing if the user specified the label
if (shape.AlreadyLabeled)
{
return;
}
// Use a particular recognizer based on how the shape was
// classified.
if (shape.Classification == LogicDomain.WIRE_CLASS)
{
_wireRecognizer.recognize(shape, featureSketch);
}
else if (shape.Classification == LogicDomain.TEXT_CLASS)
{
_textRecognizer.recognize(shape, featureSketch);
}
else if (shape.Classification == LogicDomain.GATE_CLASS)
{
_gateRecognizer.recognize(shape, featureSketch);
}
else
{
Console.WriteLine(
"Error in shape recognition: cannot recognize" +
" a shape with the classification \"" +
shape.Classification + "\"");
}
}
public virtual void process(Featurefy.FeatureSketch featureSketch)
{
foreach (Sketch.Shape shape in featureSketch.Sketch.Shapes)
{
orient(shape, featureSketch);
}
}
public void FeatureSketchCreationBehavior()
{
/*
* Problem description:
* When a sketch contains duplicated points, the creation of a FeatureSketch
* may delete those points.
*/
Sketch.Sketch sketch = Sketches.newValidSketch();
// Duplicate a point in the sketch
Sketch.Shape shape = sketch.Shapes[0];
List <Sketch.Point> points = new List <Sketch.Point>(shape.Substrokes[0].Points);
points.Add(points[points.Count - 1]);
Sketch.Substroke substroke = new Sketch.Substroke(points);
sketch.AddStroke(new Sketch.Stroke(substroke));
shape.AddSubstroke(substroke);
Sketch.Sketch clone = sketch.Clone();
Assert.IsTrue(sketch.Equals(clone), "Sketch is not equal to its clone");
Featurefy.FeatureSketch featureSketch = newValidFeatureSketch(new Sketch.Project(sketch));
Assert.IsTrue(sketch.Equals(clone), "FeatureSketch creation modified original sketch");
}
/// <summary>
/// Refines the recognition after the shapes have all been recognized.
///
/// Precondition: ConnectSketch() has been called.
/// </summary>
public void RefineSketch(Featurefy.FeatureSketch _featuresketch)
{
_refinement.process(_featuresketch);
if (debug)
{
Console.WriteLine("Connections:");
foreach (Sketch.Shape shape in _featuresketch.Sketch.Shapes)
{
Console.Write(" --> " + shape.Name + ": ");
bool first = true;
foreach (Sketch.Shape connectedShape in shape.ConnectedShapes)
{
if (!first)
{
Console.Write(", ");
}
first = false;
Console.Write(connectedShape.Name);
}
Console.WriteLine();
}
}
}
/// <summary>
/// Recognize what gate a shape is using the ComboRecgonizer,
/// and update the shape with the results.
/// </summary>
/// <param name="shape">
/// The shape to recognize (should be a gate)
/// </param>
public override void recognize(Sketch.Shape shape, Featurefy.FeatureSketch featureSketch)
{
// Initialize variables to store results in
ShapeType bestType = new ShapeType();
float bestProbability = 0F;
double bestOrientation;
// Recognize the shape and pick out the best option
Dictionary <ShapeType, double> alternativeTypes = Recognize(shape, out bestOrientation);
foreach (KeyValuePair <ShapeType, double> pair in alternativeTypes)
{
ShapeType type = pair.Key;
double prob = pair.Value;
if (prob > bestProbability)
{
bestType = type;
bestProbability = (float)prob;
}
}
// Update the shape to reflect this recognition
shape.setRecognitionResults(
bestType,
bestProbability,
alternativeTypes,
bestOrientation);
}
/// <summary>
/// Create a new worker.
/// </summary>
/// <param name="inner">the recognizer to use</param>
/// <param name="start">the first shape to recognize (inclusive)</param>
/// <param name="end">the last shape to recognize (exclusive)</param>
/// <param name="featureSketch">the sketch whose shapes we'll work on</param>
public Worker(Recognizer inner, int start, int end, Featurefy.FeatureSketch featureSketch)
{
_innerRecognizer = inner;
_start = start;
_end = end;
_featureSketch = featureSketch;
_shapes = _featureSketch.Sketch.Shapes;
}
public static InkToSketchWPF.InkCanvasSketch newInkCanvasSketch()
{
// Manually construct the necessary data members
Sketch.Sketch sketch = new Sketch.Sketch();
Featurefy.FeatureSketch featureSketch = UnitTests.FeaturefyOperations.newValidFeatureSketch(new Sketch.Project(sketch));
InkCanvas inkcanvas = new InkCanvas();
return(new InkToSketchWPF.InkCanvasSketch(inkcanvas, featureSketch));
}
/// <summary>
/// After the sketch has been recognized, we can look at the context of each shape.
/// If a shape is out of context (for example, if a NOT gate is connected to three wires),
/// then we assign the shape the next most likely label.
/// </summary>
/// <param name="featureSketch">the sketch to process</param>
public virtual void process(Featurefy.FeatureSketch featureSketch)
{
if (DEBUG)
{
Console.WriteLine("\nCareful Context Refinement");
}
foreach (Sketch.Shape shape in featureSketch.Sketch.Shapes)
{
bool valid = _domain.IsProperlyConnected(shape);
if (valid || shape.AlreadyGrouped)
{
continue;
}
ShapeType originalType = shape.Type;
string originalName = shape.Name;
float originalProbability = shape.Probability;
// If a shape only has one connection,
// it is assumed to be a label.
if (shape.ConnectedShapes.Count == 1)
{
foreach (Sketch.Substroke substroke in shape.Substrokes)
{
substroke.Classification = LogicDomain.TEXT_CLASS;
}
_sketchRecognizer.processShape(shape, featureSketch);
if (shape.Probability < 0.5)
{
// we probably made a mistake, so revert.
shape.Type = originalType;
shape.Name = originalName;
shape.Probability = originalProbability;
}
}
if (Domain.LogicDomain.IsGate(shape.Type))
{
nextBestLabel(shape);
if (!_domain.IsProperlyConnected(shape))
{
// the next best type wasn't properly connected either
// so we probably want to stick with our first type
shape.Type = originalType;
shape.Name = originalName;
shape.Probability = originalProbability;
}
}
if (DEBUG && ((originalName != shape.Name) || (originalType != shape.Type))) // if it changed
{
Console.WriteLine(" " + originalName + " (" + originalType + ") -> " + shape.Name + " (" + shape.Type + "); confidence = " + shape.Probability);
}
}
}
/// <summary>
/// Run the recognizer on a sketch.
///
/// Precondition: the sketch has been grouped into shapes.
///
/// Postcondition: shape.Type and shape.Probability have been filled in with
/// this recognizer's best guess for every shape in the sketch.
/// </summary>
/// <param name="featureSketch">the sketch to work on</param>
public void process(Featurefy.FeatureSketch featureSketch)
{
foreach (Sketch.Shape shape in featureSketch.Sketch.Shapes)
{
if (shape.AlreadyLabeled || !canRecognize(shape.Classification))
{
continue;
}
processShape(shape, featureSketch);
}
}
public void process(Featurefy.FeatureSketch featureSketch)
{
Sketch.Sketch sketch = featureSketch.Sketch;
foreach (Sketch.Shape shape in sketch.Shapes)
{
if (shape.Type != new Domain.ShapeType())
{
shape.AlreadyLabeled = true;
shape.AlreadyGrouped = true;
}
}
}
/// <summary>
/// Test the validity of each labeled shape.
/// </summary>
/// <returns>A dictionary mapping shapes to validity.</returns>
public Dictionary <Sketch.Shape, bool> TestValidity(Featurefy.FeatureSketch _featuresketch)
{
Sketch.Sketch sketch = _featuresketch.Sketch;
Dictionary <Sketch.Shape, bool> dict = new Dictionary <Sketch.Shape, bool>();
foreach (Sketch.Shape shape in sketch.Shapes)
{
dict.Add(shape, _domain.IsProperlyConnected(shape));
}
return(dict);
}
/// <summary>
/// Construct a sketch modification. Computes the benefit of this action
/// by computing the difference in the given energy function from
/// performing the given sketch operation.
/// </summary>
public SketchModification(Featurefy.FeatureSketch sketch, ISketchOperation op, Func <Featurefy.FeatureSketch, double> energyFunc)
{
_sketch = sketch.Sketch;
_op = op;
double initialEnergy = energyFunc(sketch);
op.perform();
double finalEnergy = energyFunc(sketch);
op.undo();
_benefit = finalEnergy - initialEnergy;
}
/// <summary>
/// Classifies only the given substrokes.
///
/// Postcondition: substroke.Classification is "Wire," "Gate," "Label," or "Unknown" for
/// every substroke in the list. Strokes are not reclassified if they have a parent
/// shape with AlreadyLabeled set to true.
/// </summary>
/// <param name="sketch">the sketch to use</param>
/// <param name="featureSketch">the featureSketch to use</param>
public virtual void process(Featurefy.FeatureSketch featureSketch, IEnumerable <Sketch.Substroke> substrokes)
{
foreach (Sketch.Substroke substroke in substrokes)
{
Sketch.Shape parent = substroke.ParentShape;
if (parent != null && parent.AlreadyLabeled)
{
continue;
}
substroke.Classification = classify(substroke, featureSketch);
}
}
/// <summary>
/// Classifies all the substrokes in a sketch.
///
/// Postcondition: substroke.Classification is "Wire," "Gate," "Label," or "Unknown" for
/// every substroke in the sketch. Strokes are not reclassified if they have a parent
/// shape with AlreadyLabeled set to true.
/// </summary>
/// <param name="sketch">the sketch to use</param>
public virtual void process(Featurefy.FeatureSketch sketch)
{
foreach (Sketch.Substroke substroke in sketch.Sketch.Substrokes)
{
Sketch.Shape parent = substroke.ParentShape;
if (parent != null && parent.AlreadyLabeled)
{
continue;
}
classify(substroke, sketch);
}
}
/// <summary>
/// After the sketch has been recognized, we can look at the context of each shape.
/// If a shape is out of context (for example, if a NOT gate is connected to three wires),
/// then we assign the shape the next most likely label.
/// </summary>
/// <param name="featureSketch">the sketch to process</param>
public void process(Featurefy.FeatureSketch featureSketch)
{
if (DEBUG)
{
Console.WriteLine("Context Refinement");
}
// Refine three times... (THIS IS A MAGIC NUMBER I PULLED OUT OF THE AIR).
int CONTEXT_REFINE_ITERATIONS = 3;
for (int i = 0; i < CONTEXT_REFINE_ITERATIONS; i++)
{
foreach (Sketch.Shape shape in featureSketch.Sketch.Shapes)
{
bool valid = _domain.IsProperlyConnected(shape);
if (valid || shape.AlreadyGrouped)
{
continue;
}
ShapeType originalType = shape.Type;
string originalName = shape.Name;
// If a shape only has one connection, it is assumed to be a label
if (shape.ConnectedShapes.Count == 1)
{
foreach (Sketch.Substroke substroke in shape.Substrokes)
{
substroke.Classification = "Text";
}
_sketchRecognizer.recognize(shape, featureSketch);
}
else
{
if (Domain.LogicDomain.IsGate(shape.Type))
{
nextBestLabel(shape);
}
}
if (DEBUG && ((originalName != shape.Name) || (originalType != shape.Type))) // if it changed
{
Console.WriteLine(" " + originalName + " (" + originalType + ") -> " + shape.Name + " (" + shape.Type + ")");
}
}
}
}
public void process(Featurefy.FeatureSketch featureSketch)
{
_sketch = featureSketch.Sketch;
List <Shape> notbubbles = Data.Utils.filter(_sketch.Shapes, delegate(Shape s) { return(s.Type == LogicDomain.NOTBUBBLE); });
foreach (Shape bubble in notbubbles)
{
foreach (Shape connected in bubble.ConnectedShapes)
{
if (tryToMerge(bubble, connected))
{
break;
}
}
}
}
/// <summary>
/// Regroups and recognizes the shape
/// </summary>
/// <param name="shape"></param>
private void regroupShape(Sketch.Shape shape, Featurefy.FeatureSketch _featuresketch)
{
if (shape.Type == new ShapeType())
{
_strokeClassifier.process(_featuresketch, shape.Substrokes);
}
if (shape.Substrokes.Length > 0)
{
// Determine the majority classification
Dictionary <string, int> counts = new Dictionary <string, int>();
foreach (Sketch.Substroke substroke in shape.Substrokes)
{
if (!counts.ContainsKey(substroke.Classification))
{
counts.Add(substroke.Classification, 0);
}
counts[substroke.Classification]++;
}
string bestClassification = LogicDomain.GATE_CLASS;
int bestCount = 0;
foreach (KeyValuePair <string, int> pair in counts)
{
if (pair.Value > bestCount)
{
bestCount = pair.Value;
bestClassification = pair.Key;
}
}
// Give all the substrokes the same classification since they are in the same group
shape.Classification = bestClassification;
if (!shape.AlreadyLabeled)
{
_sketchRecognizer.processShape(shape, _featuresketch);
}
}
else
{
Console.WriteLine("WARNING No substrokes available");
}
MakeShapeNamesUnique(_featuresketch);
}
public virtual void orient(Shape shape, Featurefy.FeatureSketch featureSketch)
{
ShapeType type = shape.Type;
BitmapSymbol defn = new BitmapSymbol();
foreach (BitmapSymbol template in _templates)
{
if (template.SymbolType == type)
{
defn = template;
break;
}
}
BitmapSymbol unknown = new BitmapSymbol(shape.SubstrokesL);
shape.Orientation = unknown.bestOrientation(defn);
}
/// <summary>
/// Recognizes the text a shape forms, and updates the
/// shape with the recognition results (the text and the
/// likelihood that the recognition was correct).
///
/// This method does not use the featureSketch argument.
///
/// Postcondition: the shape is of type IO and its Name
/// property tells what text it was recognized as.
/// </summary>
/// <param name="shape">The shape to recognize</param>
/// <param name="featureSketch">Not used.</param>
public override void recognize(Sketch.Shape shape, Featurefy.FeatureSketch featureSketch)
{
// Prepare the shape for recognition
SketchOrInkConverter converter = new SketchOrInkConverter();
_microsoftTextRecognizer.Strokes = converter.convertToInk(shape);
// Try to recognize the shape
RecognitionStatus status;
RecognitionResult result;
result = _microsoftTextRecognizer.Recognize(out status);
// Origanize the results
string shapeName = "";
float probability = 0F;
if ((result != null) && (status == RecognitionStatus.NoError))
{
shapeName = result.TopString;
switch (result.TopConfidence)
{
case RecognitionConfidence.Poor:
probability = .1F;
break;
case RecognitionConfidence.Intermediate:
probability = .5F;
break;
case RecognitionConfidence.Strong:
probability = .9F;
break;
}
}
// Update the shape to reflect these results
if (debug)
{
Console.WriteLine("Found input/output label: " + shapeName + " (confidence = " + probability + ")");
}
shape.setRecognitionResults(LogicDomain.TEXT, probability, shapeName);
}
/// <summary>
/// Uses the RecognitionInterfaces.Recognizer recognize method
/// which recognizes and assigns the type of a shape. This
/// implementation allows the use of the learnFromExample
/// method in Interface Functions.
/// </summary>
/// <param name="shape">The shape to recognize</param>
/// <param name="featureSketch">The featureSketch to use</param>
public override void recognize(Sketch.Shape shape, Featurefy.FeatureSketch featureSketch)
{
BitmapSymbol unknown = new BitmapSymbol(shape.SubstrokesL);
List <SymbolRank> results = unknown.Recognize(_templates);
if (results.Count > 0)
{
// Populate the dictionary of alterateTypes with all of the ShapeTypes in results
Dictionary <ShapeType, double> alternateTypes = new Dictionary <ShapeType, double>();
if (debug)
{
Console.WriteLine("\nRecognition results: ");
}
foreach (SymbolRank result in results)
{
if (!alternateTypes.ContainsKey(result.SymbolType))
{
alternateTypes.Add(result.SymbolType, getProbability(result.SymbolType, results));
}
if (debug)
{
Console.WriteLine(result.SymbolType + " with template " + result.SymbolName);
}
}
ShapeType type = results[0].SymbolType; // the most likely type
float probability = (float)alternateTypes[type]; // grab the probability of our most likely type
alternateTypes.Remove(type); // the most likely type is NOT an alternate
shape.setRecognitionResults(
type,
probability,
alternateTypes,
results[0].BestOrientation,
results[0].SymbolName);
}
}
/// <summary>
/// Create a Recognition Manager for the given sketch panel with default settings.
/// Settings are loaded from file settings.txt
/// </summary>
/// <param name="p">a sketch panel to manage</param>
public RecognitionManager(SketchPanelLib.SketchPanel p)
{
// Load settings from text file
string directory = AppDomain.CurrentDomain.BaseDirectory;
string SettingsFilename = directory + "//settings.txt";
_filenames = Files.SettingsReader.readSettings(SettingsFilename);
// Initialize the recognition machines
_domain = ContextDomain.CircuitDomain.GetInstance();
_strokeClassifier = RecognitionPipeline.createDefaultClassifier();
_strokeGrouper = RecognitionPipeline.createDefaultGrouper();
_sketchRecognizer = RecognitionPipeline.createDefaultRecognizer();
_connector = RecognitionPipeline.createDefaultConnector();
_refinement = RecognitionPipeline.createDefaultRefiner(_connector, _sketchRecognizer);
// Add events
_panel = p;
_featuresketch = p.InkSketch.FeatureSketch;
}
/// <summary>
/// Pick out the shapes that still that do not match their context,
/// try grouping neighboring strokes into them (i.e. they steal a neighboring stroke),
/// and if this puts it in context, keep it.
///
/// NOTE: we're only considering gates at the moment
/// </summary>
public void StrokeStealRefine(Featurefy.FeatureSketch featureSketch)
{
foreach (Sketch.Shape shape in featureSketch.Sketch.Shapes)
{
bool valid = _domain.IsProperlyConnected(shape);
// If the shape was not specified by the user, doesn't fit its
// context, and is a gate
if (!valid && !shape.AlreadyGrouped && shape.Classification == LogicDomain.GATE_CLASS)
{
float bestProb = 0F;
Sketch.Substroke bestSubstroke = null;
// For each neighboring shape, try stealing a stroke until
// one fits well given that the user did not specify it
List <Sketch.Shape> neighbors = featureSketch.Sketch.neighboringShapes(shape);
foreach (Sketch.Shape neighbor in neighbors)
{
if (!neighbor.AlreadyGrouped)
{
foreach (Sketch.Substroke substroke in neighbor.Substrokes)
{
float stealProb = strokeStealHelpfulness(shape, substroke, featureSketch);
if (stealProb > bestProb)
{
bestProb = stealProb;
bestSubstroke = substroke;
}
}
}
}
// Only steal the substroke that gives the best new shape
stealStroke(shape, bestSubstroke, featureSketch);
}
}
}
/// <summary>
/// Uses the cached result if it is available, or makes a call to the inner
/// recognizer otherwise. This method is thread-safe provided that
/// 1. the underlying recognizer's "recognize" method is thread-safe
/// 2. this method will not be called concurrently on the same shape
/// </summary>
/// <param name="shape">the shape to recognize</param>
/// <param name="featureSketch">the featureSketch to work on</param>
public override void recognize(Sketch.Shape shape, Featurefy.FeatureSketch featureSketch)
{
int hash = shape.GetHashCode();
// Safely determine if the value is already cached. We never remove cached
// results, so it is safe to release the lock afterward.
bool isCached;
lock (_typeResults)
isCached = _typeResults.ContainsKey(hash);
if (isCached)
{
// Write the cached results to the shape. Since we can assume that
// we don't need to lock the shape, all we need to synchronize on is
// the dictionary when we retreive results.
ShapeType type;
float prob;
lock (_typeResults)
{
type = LogicDomain.getType(_typeResults[hash]);
prob = _probabilityResults[hash];
}
shape.setRecognitionResults(type, prob);
}
else
{
// We only need a lock here when we write results to the dictionary.
_innerRecognizer.recognize(shape, featureSketch);
lock (_typeResults)
{
_typeResults.Add(hash, shape.Type.Name);
_probabilityResults.Add(hash, shape.Probability);
}
}
}
public virtual void orient(Shape shape, Featurefy.FeatureSketch featureSketch)
{
ShapeType type = shape.Type;
BitmapSymbol defn = new BitmapSymbol();
// TODO: should we use multiple templates, or just one?
foreach (BitmapSymbol template in _templates)
{
if (template.SymbolType == type)
{
defn = template;
break;
}
}
BitmapSymbol unknown = _shapesToSymbols[shape];
shape.Orientation = unknown.bestOrientation(defn);
#if JESSI
Console.WriteLine("The final shape orientation is " + shape.Orientation);
Console.WriteLine();
#endif
}
/// <summary>
/// Ensure that all the shape names in the sketch are unique.
/// </summary>
public void MakeShapeNamesUnique(Featurefy.FeatureSketch _featuresketch)
{
new Refiner.UniqueNamer().process(_featuresketch);
}
/// <summary>
/// Recognizes a shape and updates values in shape (through setRecognitionResults).
/// Also notes which template was used for template ranking purposes.
/// </summary>
/// <param name="shape">The shape to recognize</param>
/// <param name="featureSketch">The featureSketch containing the shape.
/// Never used in this function, but kept around because this function is overriding another.</param>
public override void recognize(Sketch.Shape shape, Featurefy.FeatureSketch featureSketch)
{
base.recognize(shape, featureSketch);
//++_templateUsage[shape.Type][base.findTemplate(shape.TemplateName)]; // increment the number of times that BitmapSymbol has been used
}
/// <summary>
/// Rerecognizes the strokes given as a single group
/// </summary>
public void RerecognizeGroup(Sketch.Shape shape, Featurefy.FeatureSketch _featuresketch)
{
regroupShape(shape, _featuresketch);
shape.ClearConnections();
_connector.connect(shape, _featuresketch.Sketch);
}
// It should be noted that the functions to create the ARFF files are somewhat of a hack. They work, on my computer.
// They may also work on yours. Hopefully once the ARFF files are created nobody will ever have to use these functions
// again, but we're leaving them here anyways. Use at your own risk and feel free to email me if you have questions.
// Good luck! -- Jessi Peck, Sketcher 2011, [email protected]
/// <summary>
/// Create ARFF files for the classifier and grouper from a set of sketches
/// </summary>
/// <param name="fileList">a list of sketch file names</param>
public static void createARFFs(List <string> fileList)
{
// These dictionaries will store the feature values for each of the ARFF files we're creating
Dictionary <double[], string> strokeValues = new Dictionary <double[], string>();
Dictionary <string, Dictionary <double[], string> > groupValues = new Dictionary <string, Dictionary <double[], string> >();
groupValues.Add(TEXT, new Dictionary <double[], string>());
groupValues.Add(WIRE, new Dictionary <double[], string>());
groupValues.Add(GATE, new Dictionary <double[], string>());
// These lists will store the names of the features for the ARFF files we're creating
List <string> strokeNames = new List <string>();
List <string> groupNames = new List <string>();
// Stores each stroke's classification.
Dictionary <Substroke, string> strokeClassifications = new Dictionary <Substroke, string>();
Dictionary <string, string> settingsFiles = Files.SettingsReader.readSettings();
bool firstFile = true;
foreach (string file in fileList)
{
if (!file.EndsWith(".xml"))
{
continue;
} // only read .xml files
//Console.WriteLine("Reading file " + file);
// Load the sketch from the file and make it into a feature sketch
Sketch.Sketch sketch = new ConverterXML.ReadXML(file).Sketch;
Featurefy.FeatureSketch featureSketch = Featurefy.FeatureSketch.MakeFeatureSketch(new Sketch.Project(sketch), settingsFiles);
// Get featureSketch to tell you the feature values for each individual stroke in the sketch
Dictionary <Substroke, double[]> classificationDict = new Dictionary <Substroke, double[]>();
foreach (Substroke substroke in sketch.Substrokes)
{
classificationDict.Add(substroke, featureSketch.GetValuesSingle(substroke));
}
// Classify each substroke based on what shape it's a part of
foreach (Sketch.Shape shape in sketch.Shapes)
{
string classif = shape.Type.Classification;
if (classif == UNKNOWN)
{
continue;
}
foreach (Substroke stroke in shape.Substrokes)
{
strokeValues.Add(classificationDict[stroke], classif);
strokeClassifications.Add(stroke, classif);
}
}
#region Bad XML File Catcher
// This shouldn't be necessary if your XML files are properly formatted,
// but it's here just in case.
List <Substroke> badStrokes = new List <Substroke>();
foreach (Substroke strokeToCheck in classificationDict.Keys)
{
if (classificationDict[strokeToCheck].Length != 27)
{
Console.WriteLine("OH MAN. GET VALUES SINGLE FAILS.");
Console.WriteLine("Your length of features is " + classificationDict[strokeToCheck].Length + " elements long.");
badStrokes.Add(strokeToCheck);
}
}
foreach (Substroke naughtyStroke in badStrokes)
{
strokeValues.Remove(classificationDict[naughtyStroke]);
strokeClassifications.Remove(naughtyStroke);
classificationDict.Remove(naughtyStroke);
Console.WriteLine("You will never see that stroke again.");
}
#endregion
// Get featureSketch to tell you the feature values for each pair of strokes in the sketch
Dictionary <string, Dictionary <Featurefy.FeatureStrokePair, double[]> > grouperDict = featureSketch.GetValuesPairwise(strokeClassifications);
// You only need to call this once, might as well be on the first file you do.
if (firstFile)
{
strokeNames = featureNames(featureSketch.FeatureListSingle); // set up the list of feature names for stroke classification
groupNames = featureNames(featureSketch.FeatureListPair); // set up the list of feature names for stroke grouping
firstFile = false;
}
// Adds feature values for grouping to groupValues
getGroupValues(ref groupValues, grouperDict);
}
// Populate the lists of class names
List <string> strokeClasses, groupClasses;
classNames(out strokeClasses, out groupClasses);
string directory = AppDomain.CurrentDomain.BaseDirectory;
// write ARFF for stroke classifier
writeARFFfile(directory + STROKE + ARFF, strokeNames, strokeValues, strokeClasses);
// write ARFF for stroke grouper x3
writeARFFfile(directory + TEXT_GROUP + ARFF, groupNames, groupValues[TEXT], groupClasses);
//writeARFFfile(directory + WIRE_GROUP + ARFF, groupNames, groupValues[WIRE], groupClasses);
writeARFFfile(directory + GATE_GROUP + ARFF, groupNames, groupValues[GATE], groupClasses);
//Console.WriteLine("ARFF files created!");
}
