public void LinearRegression_regression_parameters_Test01()
{
//define values, and variables
Variable x = Variable.InputVariable(new int[] { 1 }, DataType.Float, "input");
Variable y = Variable.InputVariable(new int[] { 1 }, DataType.Float, "output");
//data
var xValues = Value.CreateBatch <float>(new NDShape(1, 1), new float[] { 1f, 2f, 3f, 4f, 5f }, device);
var yValues = Value.CreateBatch <float>(new NDShape(1, 1), new float[] { 3f, 5f, 7f, 9f, 11f }, device);
//create linear regression model, LRM
NetworkFoundation net = new NetworkFoundation();
var model = (Function)net.Layer(x, 1, DataType.Float, device);
//test model by manually assign weights parameters value
var ps = model.Inputs.Where(z => z.IsParameter).ToList();
var totalParameters = ps.Sum(c => c.Shape.TotalSize);
//set bias parameter to 1
ps[0].GetValue().SetValue(1);
//set weight parameter to 2
ps[1].GetValue().SetValue(2);
//print weights
var b0_name = ps[0].Name;
var b0 = new Value(ps[0].GetValue()).GetDenseData <float>(ps[0]);
var b1_name = ps[1].Name;
var b1 = new Value(ps[1].GetValue()).GetDenseData <float>(ps[1]);
//check regression parameters
Assert.Equal(1f, b0[0][0]);
Assert.Equal(2f, b1[0][0]);
Console.WriteLine($"b0={b0[0][0]}, b1={b1[0][0]}");
//Evaluate model after weights are setup
var inV = new Dictionary <Variable, Value>();
inV.Add(x, xValues);
var outV = new Dictionary <Variable, Value>();
outV.Add(model, null);
//evaluate
model.Evaluate(inV, outV, device);
//collect result
var result = outV[model].GetDenseData <float>(model);
//check results
Assert.Equal(3f, result[0][0]);
Assert.Equal(5f, result[1][0]);
Assert.Equal(7f, result[2][0]);
Assert.Equal(9f, result[3][0]);
Assert.Equal(11f, result[4][0]);
}
public void LinearMultipleRegression_Test()
{
//define values, and variables
Variable x = Variable.InputVariable(new int[] { 3 }, DataType.Float, "input");
Variable y = Variable.InputVariable(new int[] { 1 }, DataType.Float, "output");
//data
var xData = new float[] { 1f, 2f, 3f, /*first row*/
2f, 4f, 6f, /*second row*/
3f, 6f, 9f, /*third row*/
4f, 8f, 11f, /*fourth row*/
5f, 10f, 14f /*fifth row*/ };
var xValues = Value.CreateBatch <float>(new NDShape(1, 3), xData, device);
var yValues = Value.CreateBatch <float>(new NDShape(1, 1), new float[] { 21f, 41f, 61f, 77f, 98f }, device);
//create linear regression model, LRM
NetworkFoundation net = new NetworkFoundation();
var model = (Function)net.Layer(x, 1 /**/, DataType.Float, device);
//LRM contains only two parameters b0 and b1
var ps = model.Inputs.Where(z => z.IsParameter).ToList();
var totalParameters = ps.Sum(c => c.Shape.TotalSize);
//test parameters count
Assert.Equal(4, totalParameters);
//create trainer
var trainer = createTrainer(model, y, 0.005);
//training
for (int i = 1; i <= 200; i++)
{
var d = new Dictionary <Variable, Value>();
d.Add(x, xValues);
d.Add(y, yValues);
//
trainer.TrainMinibatch(d, true, device);
//
var loss = trainer.PreviousMinibatchLossAverage();
var eval = trainer.PreviousMinibatchLossAverage();
//
Console.WriteLine($"Loss={loss}, Eval={eval}");
//print weights
var b0_name = ps[0].Name;
var b0 = new Value(ps[0].GetValue()).GetDenseData <float>(ps[0]);
var b1_name = ps[1].Name;
var b1 = new Value(ps[1].GetValue()).GetDenseData <float>(ps[1]);
if (i % 10 == 0)
{
Console.WriteLine($"b0={b0[0][0]}, b1={b1[0][0]}, b2={b1[0][1]}, b3={b1[0][2]}");
}
}
}
/// <summary>
/// Create embedding layer
/// </summary>
/// <param name="input"></param>
/// <param name="embeddingDim"></param>
/// <param name="device"></param>
/// <returns></returns>
public static Function Create(Variable x, int embeddingDim, DataType dataType, DeviceDescriptor device, uint seed, string name)
{
var f = new NetworkFoundation();
//weights
var W = f.Weights(embeddingDim, dataType, device, seed, "e");
//matrix product
var Wx = CNTKLib.Times(W, x, name);
//
return(Wx);
}
public void LinearRegression_Test()
{
//define values, and variables
Variable x = Variable.InputVariable(new int[] { 1 }, DataType.Float, "input");
Variable y = Variable.InputVariable(new int[] { 1 }, DataType.Float, "output");
//define training data set from table above
var xValues = Value.CreateBatch <float>(new NDShape(1, 1), new float[] { 1f, 2f, 3f, 4f, 5f }, device);
var yValues = Value.CreateBatch <float>(new NDShape(1, 1), new float[] { 3f, 5f, 7f, 9f, 11f }, device);
//create linear regression model, LRM
NetworkFoundation net = new NetworkFoundation();
var model = (Function)net.Layer(x, 1, DataType.Float, device);
//LRM contains only two parameters b0 and b1
var ps = model.Inputs.Where(z => z.IsParameter).ToList();
var totalParameters = ps.Sum(c => c.Shape.TotalSize);
//test parameters count
Assert.Equal(2, totalParameters);
//create trainer
var trainer = createTrainer(model, y, 0.05);
//training
for (int i = 1; i <= 100; i++)
{
var d = new Dictionary <Variable, Value>();
d.Add(x, xValues);
d.Add(y, yValues);
//
trainer.TrainMinibatch(d, true, device);
//
var loss = trainer.PreviousMinibatchLossAverage();
var eval = trainer.PreviousMinibatchLossAverage();
//
Console.WriteLine($"Loss={loss}, Eval={eval}");
//print weights
var b0_name = ps[0].Name;
var b0 = new Value(ps[0].GetValue()).GetDenseData <float>(ps[0]);
var b1_name = ps[1].Name;
var b1 = new Value(ps[1].GetValue()).GetDenseData <float>(ps[1]);
Console.WriteLine($"b0={b0[0][0]}, b1={b1[0][0]}");
}
}
