public HistogramEntry(int input)
{
Input = input;
Frequency = 1;
next = null;
}
public HistogramEntry(int input)
{
Input = input;
Frequency = 1;
next = null;
}
public static int getFrequency(HistogramEntry root, int input)
{
HistogramEntry entry = findHistogramEntry ( root, input);
int freq;
if (entry != null) freq = entry.Frequency;
else freq = 0;
return freq;
}
// update the Histogram or add a new entry
public static void increaseFrequency(ref HistogramEntry root, int input)
{
HistogramEntry entry = findHistogramEntry(root, input);
if (entry == null)
{
root = addHistogramEntry(root, input);
}
else
{
entry.Frequency += 1;
}
}
public static int HistogramLength(HistogramEntry root)
{
int length = 0;
HistogramEntry temp = root;
while (temp != null)
{
length++;
temp = temp.next;
}
return length;
}
public static HistogramEntry findHistogramEntry(HistogramEntry root, int input)
{
Boolean found = false;
HistogramEntry temp = root;
while ((!found) && (temp != null))
{
found = (temp.Input == input);
if (!found)
temp = temp.next;
}
return temp;
}
public static int HistogramLength(HistogramEntry root)
{
int length = 0;
HistogramEntry temp = root;
while (temp != null)
{
length++;
temp = temp.next;
}
return(length);
}
public static HistogramEntry findHistogramEntry(HistogramEntry root, int input)
{
Boolean found = false;
HistogramEntry temp = root;
while ((!found) && (temp != null))
{
found = (temp.Input == input);
if (!found)
{
temp = temp.next;
}
}
return(temp);
}
public static int getFrequency(HistogramEntry root, int input)
{
HistogramEntry entry = findHistogramEntry(root, input);
int freq;
if (entry != null)
{
freq = entry.Frequency;
}
else
{
freq = 0;
}
return(freq);
}
// a constantly populated histogram (aiding the ongoing clustering process)
//public int[] InputHistogram;
public STANN(int inputnum, int inputrange, int layernum,
int neuronnum, int outputnum, double threshold,
double lr, Random rand, Boolean forcedSelfTrain)
: base(inputnum, layernum, neuronnum, outputnum, threshold, lr, rand)
{
ForcedSelfTrain = forcedSelfTrain;
InputRange = inputrange;
if (ForcedSelfTrain)
{
// computing number of clusters
ClusterNum = (int)Math.Pow(2, OutputNum);
// initializing the centroids array and the input sums per cluster
resetClusterCentroids();
// retrieving the number of possible input permutations
InputPermutationsNum = getInputPermsNumber();
// resetting the Input Histogram
InputHisto = null;
//resetInputHistogram();
}
}
// a constantly populated histogram (aiding the ongoing clustering process)
//public int[] InputHistogram;
public STANN(int inputnum, int inputrange, int layernum,
int neuronnum, int outputnum, double threshold,
double lr, Random rand, Boolean forcedSelfTrain)
: base(inputnum, layernum, neuronnum, outputnum, threshold, lr, rand)
{
ForcedSelfTrain = forcedSelfTrain;
InputRange = inputrange;
if (ForcedSelfTrain)
{
// computing number of clusters
ClusterNum = (int)Math.Pow(2, OutputNum);
// initializing the centroids array and the input sums per cluster
resetClusterCentroids();
// retrieving the number of possible input permutations
InputPermutationsNum = getInputPermsNumber();
// resetting the Input Histogram
InputHisto = null;
//resetInputHistogram();
}
}
public static HistogramEntry addHistogramEntry(HistogramEntry root, int input)
{
HistogramEntry temp = root;
HistogramEntry newroot;
if (temp == null)
{
temp = new HistogramEntry(input);
newroot = temp;
}
else
{
newroot = root;
while (temp.next != null)
temp = temp.next;
temp.next = new HistogramEntry(input);
}
return newroot;
}
public static HistogramEntry addHistogramEntry(HistogramEntry root, int input)
{
HistogramEntry temp = root;
HistogramEntry newroot;
if (temp == null)
{
temp = new HistogramEntry(input);
newroot = temp;
}
else
{
newroot = root;
while (temp.next != null)
{
temp = temp.next;
}
temp.next = new HistogramEntry(input);
}
return(newroot);
}
/*public int histoElementNumber()
* {
* int i, count = 0;
* for (i = 0; i < InputPermutationsNum; i++)
* count += InputHistogram[i];
* return count;
* }*/
/* public void autoClustering1(double[] inputvec)
* {
* int i, j;
* int inputHistoIndex;
* Boolean equal;
*
* // updating the histogram
* inputHistoIndex = getInputVectorIndex(inputvec);
* InputHistogram[inputHistoIndex]++;
* int vectorPoolLength = histoElementNumber();
* // histogram updated
*
* // creating a temporary array to store centroids calculated in the previous iteration step
* double[][] prevCentroids = new double[ClusterNum][];
*
* for (i = 0; i < ClusterNum; i++)
* prevCentroids[i] = new double[InputNum];
* // array created and initialized
*
* // iterating
* do
* {
* // copying centroids to previous centrodis array
* copyCentroids(ref prevCentroids, ClusterCentroids);
* // resetting the sums
* resetClusterSums();
* // reseting the vector count for each cluster
* resetClusterVectorCount();
* // creating sums using the histogram
* for (i = 0; i < InputPermutationsNum; i++)
* {
* if (InputHistogram[i] > 0)
* {
* double[] thevec = getHistogramVector(i);
* // now finding the closest centroid to the vector at hand
* double minDistance = getEuclideanDistance(ClusterCentroids[0], thevec, InputNum);
* int minDistanceCluster = 0;
* for (j = 0; j < ClusterNum; j++)
* {
* double dist = getEuclideanDistance(ClusterCentroids[j], thevec, InputNum);
* if (dist <= minDistance)
* {
* minDistance = dist;
* minDistanceCluster = j;
* }
* }
* // closest centroid found.
* // Adding the vector to the sum weighted by the histogram entry
* int k;
* for (k = 0; k < InputNum; k++)
* ClusterSums[minDistanceCluster][k] += InputHistogram[i] * thevec[k];
* // vector added (f-times, f is the histogram entry for this particular vector)
* // increasing the vector count of the cluster
* ClusterVectorCount[minDistanceCluster] += InputHistogram[i];
* }
* }
*
* // finding new centroids (computing averages)
* for (i = 0; i < ClusterNum; i++)
* if (ClusterVectorCount[i] > 0)
* for (j = 0; j < InputNum; j++)
* ClusterCentroids[i][j] = ClusterSums[i][j] / ClusterVectorCount[i];
* // new centroids stored
* // Now checking if previous centroids are equal to the new centroids
* equal = true;
* for (i = 0; i < ClusterNum; i++)
* {
* if (!equal) break;
* for (j = 0; j < InputNum; j++)
* if (prevCentroids[i][j] != ClusterCentroids[i][j])
* {
* equal = false;
* break;
* }
* }
* } while (!equal);
* }
*/
public void autoClustering(double[] inputvec)
{
int i, j;
int inputHistoIndex;
Boolean equal;
// updating the histogram
inputHistoIndex = mapVector2Int(inputvec, InputRange, InputNum);
HistogramEntry.increaseFrequency(ref InputHisto, inputHistoIndex);
int vectorPoolLength = HistogramEntry.HistogramLength(InputHisto);
// histogram updated
// creating a temporary array to store centroids calculated in the previous iteration step
double[][] prevCentroids = new double[ClusterNum][];
for (i = 0; i < ClusterNum; i++)
{
prevCentroids[i] = new double[InputNum];
}
// array created and initialized
// iterating
do
{
// copying centroids to previous centrodis array
copyCentroids(ref prevCentroids, ClusterCentroids);
// resetting the sums
resetClusterSums();
// reseting the vector count for each cluster
resetClusterVectorCount();
// creating sums using the histogram
HistogramEntry temp;
for (temp = InputHisto; temp != null; temp = temp.next)
{
int[] intvec = mapInt2Vector(temp.Input, InputRange, InputNum);
double[] thevec = new double[InputNum];
for (int c = 0; c < InputNum; c++)
{
thevec[c] = (double)intvec[c];
}
// now finding the closest centroid to the vector at hand
double minDistance = getEuclideanDistance(ClusterCentroids[0], thevec, InputNum);
int minDistanceCluster = 0;
for (j = 0; j < ClusterNum; j++)
{
double dist = getEuclideanDistance(ClusterCentroids[j], thevec, InputNum);
if (dist <= minDistance)
{
minDistance = dist;
minDistanceCluster = j;
}
}
// closest centroid found.
// Adding the vector to the sum weighted by the histogram entry
int k;
for (k = 0; k < InputNum; k++)
{
ClusterSums[minDistanceCluster][k] += temp.Frequency * thevec[k];
}
// vector added (f-times, f is the histogram entry for this particular vector)
// increasing the vector count of the cluster
ClusterVectorCount[minDistanceCluster] += temp.Frequency;
}
// finding new centroids (computing averages)
for (i = 0; i < ClusterNum; i++)
{
if (ClusterVectorCount[i] > 0)
{
for (j = 0; j < InputNum; j++)
{
ClusterCentroids[i][j] = ClusterSums[i][j] / ClusterVectorCount[i];
}
}
}
// new centroids stored
// Now checking if previous centroids are equal to the new centroids
equal = true;
for (i = 0; i < ClusterNum; i++)
{
if (!equal)
{
break;
}
for (j = 0; j < InputNum; j++)
{
if (prevCentroids[i][j] != ClusterCentroids[i][j])
{
equal = false;
break;
}
}
}
} while (!equal);
}
// update the Histogram or add a new entry
public static void increaseFrequency(ref HistogramEntry root, int input)
{
HistogramEntry entry = findHistogramEntry (root, input);
if (entry == null)
root = addHistogramEntry (root, input);
else
entry.Frequency += 1;
}
