public void PushN(float[] Elements)
{
DisposableFloatParallelArray FPAElem = new DisposableFloatParallelArray(Elements);
if (IsEmpty)
{
m_ContArray = FPAElem;
IsEmpty = false;
}
else
{
m_ContArray = ParallelArraysUtils.Append(m_ContArray, FPAElem, 0);
m_ContArray = ParallelArrays.Evaluate(m_ContArray);
}
}
public override void Init(KohonenSOM parent)
{
this.m_Parent = parent;
float[,] globalInput = new float[m_Parent.DataSource.GetPatternCount(), m_Parent.DataSource.GetPatternLength()];
for (int i = 0; i < m_Parent.DataSource.PatternCount; ++i)
{
for (int j = 0; j < m_Parent.DataSource.GetPatternLength(); ++j)
{
globalInput[i, j] = m_Parent.DataSource.GetPattern(i)[j];
}
}
m_GPUInput = new DisposableFloatParallelArray(globalInput);
m_GPUWeight = new DisposableFloatParallelArray(m_Parent.NeuronMap);
m_GPUCoord = new DisposableFloatParallelArray(m_Parent.NeuronMapCoordArray);
}
public override void DoEpoch(float t, float round_t)
{
float[,] test2d;
float[] test;
this.FindBMU();
//Slice the pwinner row by row and do some great stuff
m_PWinner = ParallelArrays.Evaluate(m_PWinner);
Slice[] slices = new Slice[2];
for (int i = 0; i < m_Parent.DataSource.PatternCount; ++i)
{
slices[1] = new Slice(0, m_Parent.NeuronMap.GetLength(0));
slices[0] = new Slice(i, 1);
FloatParallelArray s = ParallelArrays.Section(m_PWinner, slices);
s = ParallelArrays.Evaluate(s);
FloatParallelArray bmuw = ParallelArrays.DropDimension(ParallelArrays.InnerProduct(s, m_GPUWeight), 0);
FloatParallelArray bmuc = ParallelArrays.InnerProduct(s, m_GPUCoord);
//Compute distances to bmu
DisposableFloatParallelArray bmucEvaluated = ParallelArrays.Evaluate(bmuc); //Workaround
bmuc = ParallelArrays.Stretch(bmucEvaluated, m_Parent.NeuronMap.GetLength(0), 1);
FloatParallelArray diff = ParallelArrays.Subtract(m_GPUCoord, bmuc);
FloatParallelArray dist = ParallelArrays.Multiply(diff, diff);
dist = ParallelArrays.Sum(dist, 1);
dist = ParallelArrays.Multiply(-1.0f, dist);
//Apply update formula
FloatParallelArray constE = new FloatParallelArray((float)(Math.E), m_Parent.NeuronMap.GetLength(0));
FloatParallelArray sigma = new FloatParallelArray((float)(Math.Pow(Neighborhood(t, round_t) * 0.85, 2)), m_Parent.NeuronMap.GetLength(0));
FloatParallelArray lrate = new FloatParallelArray((float)LearningRate(t, round_t), m_Parent.NeuronMap.GetLength(0), m_Parent.DataSource.PatternLength);
FloatParallelArray omeg = ParallelArrays.Divide(dist, sigma);
//FloatParallelArray momeg = ParallelArrays.Pow(constE, omeg);
FloatParallelArray momeg = ParallelArrays.Pow2(ParallelArrays.Log2(constE) * omeg);
DisposableFloatParallelArray domeg = ParallelArrays.Evaluate(momeg); //Workaround
omeg = ParallelArrays.AddDimension(domeg, 1);
omeg = ParallelArrays.Stretch(omeg, 1, m_Parent.DataSource.PatternLength);
FloatParallelArray sbmuw = ParallelArrays.AddDimension(bmuw, 0);
sbmuw = ParallelArrays.Stretch(sbmuw, m_Parent.NeuronMap.GetLength(0), 1);
m_GPUWeight = ((m_GPUWeight + ((sbmuw - m_GPUWeight) * omeg * lrate)));
}
m_GPUWeight = ParallelArrays.Evaluate(m_GPUWeight);
}
public override void FindBMU()
{
//Normalize the weight vector
FloatParallelArray transpose = ParallelArrays.Transpose(m_GPUWeight, 1, 0);
FloatParallelArray weightsq = ParallelArrays.InnerProduct(m_GPUWeight, ParallelArrays.Transpose(m_GPUWeight, 1, 0));
FloatParallelArray weightsum = ParallelArrays.Sum(weightsq, 0);
FloatParallelArray weightlength = ParallelArrays.Sqrt(weightsum);
weightlength = ParallelArrays.Stretch(ParallelArrays.AddDimension(weightlength, 1), 1, m_Parent.DataSource.PatternLength);
FloatParallelArray weightnorm = ParallelArrays.Divide(m_GPUWeight, weightlength);
weightnorm = ParallelArrays.Transpose(weightnorm, 1, 0);
//Normalize the input vector
FloatParallelArray inputsq = ParallelArrays.InnerProduct(m_GPUInput, ParallelArrays.Transpose(m_GPUInput, 1, 0));
FloatParallelArray inputsum = ParallelArrays.Sum(inputsq, 0);
FloatParallelArray inputlength = ParallelArrays.Sqrt(inputsum);
inputlength = ParallelArrays.Stretch(ParallelArrays.AddDimension(inputlength, 1), 1, m_Parent.DataSource.PatternLength);
FloatParallelArray inputnorm = ParallelArrays.Divide(m_GPUInput, inputlength);
FloatParallelArray pacc = ParallelArrays.InnerProduct(inputnorm, weightnorm);
//Replication bug here...
FloatParallelArray bmxval = ParallelArrays.MaxVal(pacc, 1);
//MSR Vivian Swelson workaround
DisposableFloatParallelArray bmxvalEvaluated = ParallelArrays.Evaluate(bmxval);
bmxval = ParallelArrays.AddDimension(bmxvalEvaluated, 1);
bmxval = ParallelArrays.Stretch(bmxval, 1, m_Parent.NeuronMap.GetLength(0));
//Winner matrix (0 = winner)
FloatParallelArray pwinner = ParallelArrays.Subtract(pacc, bmxval);
//Convert to 1 = winner, 0 otherwise
FloatParallelArray zero = new FloatParallelArray(0.0f, pwinner.Shape);
FloatParallelArray one = new FloatParallelArray(1.0f, pwinner.Shape);
BoolParallelArray bmask = ParallelArrays.CompareEqual(pwinner, zero);
m_PWinner = ParallelArrays.Cond(bmask, one, zero);
}
public override void Init(KohonenSOM parent)
{
this.m_Parent = parent;
float[,] globalInput = new float[m_Parent.DataSource.GetPatternCount(), m_Parent.DataSource.GetPatternLength()];
for (int i = 0; i < m_Parent.DataSource.PatternCount; ++i)
for (int j = 0; j < m_Parent.DataSource.GetPatternLength(); ++j)
{
globalInput[i, j] = m_Parent.DataSource.GetPattern(i)[j];
}
m_GPUInput = new DisposableFloatParallelArray(globalInput);
m_GPUWeight = new DisposableFloatParallelArray(m_Parent.NeuronMap);
m_GPUCoord = new DisposableFloatParallelArray(m_Parent.NeuronMapCoordArray);
}
public void PushN(float[] Elements)
{
DisposableFloatParallelArray FPAElem = new DisposableFloatParallelArray(Elements);
if (IsEmpty)
{
m_ContArray = FPAElem;
IsEmpty = false;
}
else
{
m_ContArray = ParallelArraysUtils.Append(m_ContArray, FPAElem, 0);
m_ContArray = ParallelArrays.Evaluate(m_ContArray);
}
}
