public virtual unsafe void Minimize()
{
DeleteTerms();
PreCheck();
Index a = new Index(OuterShape);
a.SetZero();
for (int i = 0; i < OuterShape.TotalSize; i++, a.Increase(1))
{
GetTerm(a);
}
if (Terms.Count > 1)
{
Terms.Add min = new Terms.Add(Terms.ToArray());
min.Minimize();
min.DeleteResults();
}
else if (Terms.Count == 1)
{
Terms[0].Minimize();
}
DeleteTerms();
}
public static unsafe void bb6()
{
Variable v = new Variable(new Dimension[] { 6 }, new Shape(3));
DynamicRecurrent r = new DynamicRecurrent(v.OuterDimensions, v.InnerDimensions, new[] { v },
(Layer me, List <Layer> x, Index t) =>
{
if (t[0] % 2 == 1)
{
return(new Terms.Add(x[0].GetTerm(t), me.GetTerm(t - 1)));
}
else
{
return(x[0].GetTerm(t));
}
});
r.PreCheck();
Index a = new Index(r.OuterShape);
a.SetZero();
for (int i = 0; i < r.OuterShape.TotalSize; i++, a.Increase(1))
{
Console.WriteLine(r.GetTerm(a).GetResult());
}
}
public unsafe void MultiplyTerm()
{
Input x = new Input(4, 2, 1);
Tensor data = new Tensor((10, 3, 4), DeviceConfig.Host_Float);
for (int i = 0; i < data.Shape.TotalSize; i++)
{
((float *)data.Array)[i] = i / 12;
}
x.SetInput(data);
Index a = new Index(x.OuterShape);
a.SetZero();
for (int i = 0; i < x.OuterShape.TotalSize; i++, a.Increase(1))
{
Term t = x.GetTerm(a);
Console.WriteLine("Term " + i + ": " + x.GetTerm(a).GetResult());
Term mul = new Multiply(t, t);
Console.WriteLine("Term " + i + ": " + mul.GetResult());
mul.DeleteResults();
//todo check the result.
}
}
public static unsafe void bb5()
{
Variable v = new Variable(new[] { new Dimension(3) }, new Shape(1000));
v.PreCheck();
Index a = new Index(v.OuterShape);
a.SetZero();
Terms.ReLU r = new Terms.ReLU(v.GetTerm(a));
//Console.WriteLine(v.GetTerm(a).GetResult());
//Console.WriteLine(r.GetResult());
r.DeleteResults();
((Terms.Variable)v.GetTerm(a)).Clean();
}
public unsafe override void CalculateDerivate(Tensor s)
{
if (Terms[0].ContainsTrainable) //todo kernelize this!
{
Tensor combined = new Tensor(Terms[0].Shape.Clone(), DeviceConfig.Host_Float);
float *ptrcombined = (float *)combined.Array;
float *ptrs = (float *)s.Array;
Index iterator = new Index(this.Terms[0].Shape);
iterator.SetZero();
for (int h = 0; h < this.Terms[0].Shape.TotalSize; h++)
{
int indexs = 0;
for (int i = iterator.N - 1; i >= 0; i--)
{
if (iterator.Indices[i] == this.Terms[0].Shape[i])
{
iterator.Indices[i] = 0;
iterator.Indices[i - 1]++;
}
indexs += (iterator.Indices[i] / Divisor[i]) * this.Shape.Multiplied[i + 1];
}
ptrcombined[h] = ptrs[indexs];
iterator.Indices[iterator.N - 1]++;
}
//for (int i1 = 0; i1 < s.D1; i1++)
// for (int i2 = 0; i2 < s.D2; i2++)
// for (int i3 = 0; i3 < v1d1; i3++)
// for (int i4 = 0; i4 < v1d2; i4++)
// combined[i1, i2, i3, i4] += s[i1, i2, i3 / RowDivider, i4 / ColumnDivider];// * (m[i3 / RowDivider, i4 / ColumnDivider, i3, i4] = 1);
Terms[0].Derivate(combined);
combined.Dispose();
}
}
public static void ExpandFloat32_GetGradient_0(Tensor res, Tensor s, Shape thisShape, Shape term0, Shape Multiplier)
{
res.SetValue(0);
float *ptrcombined = (float *)res.Base.Array;
float *ptrs = (float *)s.Base.Array;
if (Multiplier.N == 2 && Multiplier[1] == 1)
{
for (int i = 0; i < Multiplier[0]; i++)
{
float *me = ((float *)s.Base.Array) + i * term0.TotalSize;
VectorizationFloat.ElementWiseAddAVX((float *)res.Base.Array, me, (float *)res.Base.Array, term0.TotalSize);
}
}
else
{
Index iterator = new Index(thisShape);
iterator.SetZero();
for (int h = 0; h < thisShape.TotalSize; h++)
{
long indexs = 0;
for (int i = iterator.N - 1; i >= 0; i--)
{
if (iterator.Indices[i] == thisShape[i])
{
iterator.Indices[i] = 0;
iterator.Indices[i - 1]++;
}
indexs += (iterator.Indices[i] / Multiplier[i]) * term0.Multiplied[i + 1];
}
ptrcombined[indexs] += ptrs[h];
iterator.Indices[iterator.N - 1]++;
}
}
}
public static unsafe void bb4()
{
Input x = new Input(4, 2, 1);
var sum = new ShiftTime(x, new Dimension[] { -1 }); sum.Name = "sum";
Stopwatch c = new Stopwatch();
c.Start();
for (int i2 = 0; i2 < 1; i2++)
{
Tensor data = new Tensor((10, 3, 4), DeviceConfig.Host_Float);
for (int i = 0; i < data.Shape.TotalSize; i++)
{
((float *)data.Array)[i] = i / 12;
}
x.SetInput(data);
sum.PreCheck();
Index a = new Index(x.OuterShape);
a.SetZero();
for (int i = 0; i < x.OuterShape.TotalSize; i++, a.Increase(1))
{
Console.WriteLine("Term " + i + ":" + sum.GetTerm(a).GetResult());
}
c.Restart();
sum.Minimize();
c.Stop();
data.Dispose();
Console.WriteLine($"{i2} took {c.ElapsedMilliseconds}ms");
}
}
public unsafe static void XORExample()
{
//Hyperparameters
Hyperparameters.LearningRate = 0.1f;
Hyperparameters.Optimizer = new SGD();
//Model Creation
var l1 = LayerBuilder.Dense(16, "sigmoid");
var l2 = LayerBuilder.Dense(1, "sigmoid")[l1];
var x = new Input(2);
Layer model = l2[x];
//Loss Function Creation
var y = new Input(1);
var loss = LayerBuilder.SquaredError(model, y);
//Data preparation
Tensor x_train = new Tensor((1, 4, 2), DeviceConfig.Host_Float);
Tensor y_train = new Tensor((1, 4, 1), DeviceConfig.Host_Float);
float *xt = (float *)x_train.Array;
float *yt = (float *)y_train.Array;
// 1,1 = 0
// 1,0 = 1
// 0,1 = 1
// 0,0 = 0
xt[0] = 1; xt[1] = 1;
xt[2] = 1; xt[3] = 0;
xt[4] = 0; xt[5] = 1;
xt[6] = 0; xt[7] = 0;
yt[0] = 0;
yt[1] = 1;
yt[2] = 1;
yt[3] = 0;
//Give data to the model
x.SetInput(x_train);
y.SetInput(y_train);
Stopwatch s = new Stopwatch();
s.Start();
//Minimizing
loss.PreCheck();
Index a = new Index(model.OuterShape);
a.SetZero();
for (int epoch = 0; epoch < 100000; epoch++)
{
loss.Minimize();
if (epoch % 5000 == 0)
{
float res = ((float *)loss.GetTerm(a).GetResult().Array)[0];
res += ((float *)loss.GetTerm(a).GetResult().Array)[1];
res += ((float *)loss.GetTerm(a).GetResult().Array)[2];
res += ((float *)loss.GetTerm(a).GetResult().Array)[3];
Console.WriteLine(res);
}
}
s.Stop();
Console.WriteLine("Time Elapsed: " + s.ElapsedMilliseconds);
//Print Pools
PrintPools();
//Print the results
var result = model.GetTerm(a).GetResult();
Console.WriteLine("Results: " + result);
//Print the results of clone model
Input x2 = new Input(2);
x2.SetInput(x_train);
var clonemodel = l2[x2];
clonemodel.PreCheck();
var result2 = clonemodel.GetTerm(a).GetResult();
Console.WriteLine("Results: " + result2);
clonemodel.DeleteTerms();
model.DeleteTerms();
}
public unsafe static void MNISTExample()
{
//Hyperparameters
Hyperparameters.LearningRate = 0.001f;
Hyperparameters.Optimizer = new SGD();
//Model Creation
var x = new Input(784);
//var dropout = new Dropout(x, 0.1f);
//var model = LayerBuilder.Dense(500, x, "relu");
var model = LayerBuilder.Dense(100, x, "relu");
model = LayerBuilder.Dense(400, model, "relu");
model = LayerBuilder.Dense(200, model, "relu");
model = LayerBuilder.Dense(100, model, "relu");
model = LayerBuilder.Dense(10, model, "softmax");
//Loss Function Creation
var y = new Input(10);
var loss = LayerBuilder.SquaredError(model, y);
//Data preparation
(float[,] traindata, float[,] labels) = LoadMNISTDataSet();
int mnistsize = 42000;
Tensor x_train = Tensor.LoadArrayToDisposedTensor(traindata, new Shape(mnistsize, 784), DeviceConfig.Host_Float);
Tensor y_train = Tensor.LoadArrayToDisposedTensor(labels, new Shape(mnistsize, 10), DeviceConfig.Host_Float);
//Training
int batchsize = 100;
int trainl = 41000;
Stopwatch s = new Stopwatch();
for (int epoch = 0; epoch < 35; epoch++)
{
float l = 0;
float val = 0;
s.Restart();
Console.WriteLine("Epoch " + epoch + " başladı.");
for (int batch = 0; batch < trainl / batchsize; batch++)
{
Tensor batchx = Tensor.Cut(x_train, batch * (batchsize * 784), new Shape(1, batchsize, 784));
Tensor batchy = Tensor.Cut(y_train, batch * (batchsize * 10), new Shape(1, batchsize, 10));
x.SetInput(batchx);
y.SetInput(batchy);
loss.Minimize();
Index zero = new Index(loss.OuterShape);
zero.SetZero();
Tensor res = loss.GetTerm(zero).GetResult();
float *pp = (float *)res.Array;
for (int i = 0; i < res.Shape.TotalSize; i++)
{
l += pp[i];
}
}
for (int batch = trainl / batchsize; batch < mnistsize / batchsize; batch++)
{
Tensor batchx = Tensor.Cut(x_train, batch * (batchsize * 784), new Shape(1, batchsize, 784));
Tensor batchy = Tensor.Cut(y_train, batch * (batchsize * 10), new Shape(1, batchsize, 10));
model.DeleteTerms();
x.SetInput(batchx);
y.SetInput(batchy);
Index zero = new Index(model.OuterShape);
zero.SetZero();
model.PreCheck();
Tensor res = model.GetTerm(zero).GetResult();
for (int i = 0; i < batchsize; i++)
{
int myans = MaxId((float *)res.Array + i * 10);
int correctres = MaxId((float *)batchy.Array + i * 10);
val += (myans == correctres ? 1 : 0);
}
}
s.Stop();
Console.WriteLine("Epoch " + epoch + " biti.");
Console.WriteLine("Loss: " + l / trainl);
Console.WriteLine("Validation: " + val / (mnistsize - trainl));
Console.WriteLine("Time: " + s.ElapsedMilliseconds + "ms");
}
PrintPools();
while (true)
{
try
{
float[] data = LoadCurrentImage();
Tensor x_test = Tensor.LoadArrayToDisposedTensor(data, new Shape(1, 1, 784), DeviceConfig.Host_Float);
model.DeleteTerms();
x.SetInput(x_test);
Index zero = new Index(model.OuterShape);
zero.SetZero();
model.PreCheck();
Tensor res = model.GetTerm(zero).GetResult();
Console.WriteLine("Result: " + res);
Console.WriteLine("Digit Prediction: " + MaxId((float *)res.Array));
Console.WriteLine("-----------");
}
catch (Exception)
{
}
Thread.Sleep(500);
}
}