public int recallNetworkGuess(OCRCharacter character)
{
// Insert the 64 bit OCR data for the image as inputs.
double[] imgData = character.getPoints();
for (int i = 0; i < imgData.Length; i++)
{
inputNeuronsArr[i] = (float)imgData[i];
}
// Calculate the network
calculateNetwork();
float winner = -1;
int index = 0;
// Find the best fitting output
for (int i = 0; i < outputs; i++)
{
if (outputNeuronsArr[i] > winner)
{
winner = outputNeuronsArr[i];
index = i;
}
}
return(index);
}
public double getDistance(IDistance distCalculator, OCRCharacter otherCharacter)
{
double dist = distCalculator.getDistance(this.pointsArr, otherCharacter.getPoints());
distance = dist;
return(dist);
}
private static List <OCRCharacter> findKNearestNeighbors(
List <OCRCharacter> trainListCharacters,
IDistance calculator, OCRCharacter currentCharacter, int kSize)
{
trainListCharacters.Sort(delegate(OCRCharacter itemA, OCRCharacter itemB)
{
return(itemA.getDistance(calculator, currentCharacter).CompareTo(itemB.getDistance(calculator, currentCharacter)));
});
List <OCRCharacter> listRet = trainListCharacters.Take <OCRCharacter>(kSize).ToList();
return(listRet);
}
/**
* Given the 64 bit OCR data for binarised image and the
* expected outcome value, tests the network for a correct output.
*
* Runs the 64 bit inputs through the network and checks
* which output it gets.
*
* @param character: OCRCharacter obj with 64 bit numeric inputs
* representing the OCR data & identifier.
* @return good or bad (true or false).
*/
public bool recallNetwork(OCRCharacter character)
{
// Populate the input neurons. //
int answer = character.getIdentifier();
int index = recallNetworkGuess(character);
bool correct = index == answer ? true : false;
// Output feedback
Console.WriteLine("Input data for: " + answer + " / output: "
+ index + " (" + answer + ")" + " correct: " + correct);
return(correct);
}
public static int processNNAndPredict(List <OCRCharacter> charactersTrain,
OCRCharacter characterTest, IDistance distCalc)
{
try
{
int answer = int.MinValue;
answer = processAndPredict(characterTest, charactersTrain, distCalc);
return(answer);
}
catch (Exception e)
{
throw e;
}
}
/**
* Runs the KNN classifier on the given a character and a training set.
*
* @param charactersTrain: Training set array.
* @param charactersTest: Test set array.
* @param distCalc: Distance metric to use ex: Euclidean, Manhattan etc.
* @param kSize : The number of neoighbours to choose for classification vote.
* @param useDistanceScore: True use scoring based on euclidean distance for
* classification, False uses majority votes.
*
*/
public static int processKNNAndPredict(List <OCRCharacter> charactersTrain,
OCRCharacter characterTest, IDistance distCalc, int kSize,
bool useDistanceScore)
{
try
{
List <OCRCharacter> nearestNeighbours =
findKNearestNeighbors(charactersTrain,
distCalc, characterTest, kSize);
int classLabel = classify(nearestNeighbours, useDistanceScore);
return(classLabel);
}
catch (Exception e)
{
Console.WriteLine("ERROR " + e.Message);
return(-1);
}
}
private static int processAndPredict(OCRCharacter testChar,
List <OCRCharacter> charactersTrainList, IDistance distCalc)
{
double minDistance = 0;
int minCharIdentifier = Int32.MinValue;
for (int i = 0; i < charactersTrainList.Count; i++)
{
// Compare distance between vectors array using a distance metric (ex: EuclideanDistance distance).
double distance = charactersTrainList[i].getDistance(distCalc, testChar);
// If got a shorter distance than before, make this the current nearest distance.
if (minDistance == 0 || distance < minDistance)
{
minDistance = distance;
minCharIdentifier = charactersTrainList[i].getIdentifier();
}
}
return(minCharIdentifier);
}
/**
* Given a a list of neighbouring characters, checks which character kind has
* the most frequent occurrence (majority)
* @param neighbors
* @param useDistanceScore: True use scoring based on euclidean distance for
* classification, False uses majority votes.
* @return charIdentifier: character which has highest majority.
*/
private static int classify(List <OCRCharacter> neighbors, bool useDistanceScore)
{
Dictionary <int, double> charOccurenceList = new Dictionary <int, double>();
int num = neighbors.Count;
for (int index = 0; index < num; index++)
{
OCRCharacter temp = neighbors[index];
int key = temp.getIdentifier();
// If character kind has not been added yet.
if (!charOccurenceList.ContainsKey(key))
{
// Best practise is to add the total distance of the neighbour char type as score value.
if (useDistanceScore)
{
charOccurenceList.Add(key, 1 / temp.distance);
}
else
{
charOccurenceList.Add(key, 1.0);
}
}
// If already added, add occurence majority vote
else
{
double value = charOccurenceList[key];
// Best practise is to add the total distance of the neighbour char type as score value.
if (useDistanceScore)
{
value += 1 / temp.distance;
}
else
{
value += 1.0;
}
charOccurenceList[key] = value; // update existing value + 1
}
}
// Check which kind of identifiers have the most of the same type
double majorityScore = 0;
int charIdentifier = -1;
// Loop and determine which char identifier scored the highest
foreach (KeyValuePair <int, double> entry in charOccurenceList)
{
int identifier = entry.Key;
double value = entry.Value;
if (value > majorityScore)
{
majorityScore = value;
charIdentifier = identifier;
}
}
return(charIdentifier);
}
