static Model CreateModel(Function inputVariable, Variable targetVariable, int targetCount,
DataType dataType, DeviceDescriptor device)
{
var random = new Random(232);
Func <CNTKDictionary> weightInit = () => Initializers.GlorotNormal(random.Next());
var biasInit = Initializers.Zero();
// Create the architecture.
var network = inputVariable
.Dense(32, weightInit(), biasInit, device, dataType)
.ReLU()
.Dense(32, weightInit(), biasInit, device, dataType)
.ReLU()
.Dense(targetCount, weightInit(), biasInit, device, dataType);
// loss
var lossFunc = Losses.MeanSquaredError(network.Output, targetVariable);
var metricFunc = Losses.MeanAbsoluteError(network.Output, targetVariable);
// setup trainer.
var learner = CntkCatalyst.Learners.Adam(network.Parameters());
var trainer = CNTKLib.CreateTrainer(network, lossFunc, metricFunc, new LearnerVector {
learner
});
var model = new Model(trainer, network, dataType, device);
Trace.WriteLine(model.Summary());
return(model);
}
public void MeanSquareError_Zero_Error()
{
var targetsData = new float[] { 0, 0, 0, 0, 0, 0 };
var targetsVariable = CNTKLib.InputVariable(new int[] { targetsData.Length }, m_dataType);
var predictionsData = new float[] { 0, 0, 0, 0, 0, 0 };
var predictionsVariable = CNTKLib.InputVariable(new int[] { predictionsData.Length }, m_dataType);
var sut = Losses.MeanSquaredError(predictionsVariable, targetsVariable);
var actual = Evaluate(sut, targetsVariable, targetsData,
predictionsVariable, predictionsData);
Assert.AreEqual(0.0f, actual);
}
public void MeanSquareError()
{
var targetsData = new float[] { 1.0f, 2.3f, 3.1f, 4.4f, 5.8f };
var targetsVariable = CNTKLib.InputVariable(new int[] { targetsData.Length }, m_dataType);
var predictionsData = new float[] { 1.0f, 2.0f, 3.0f, 4.0f, 5.0f };
var predictionsVariable = CNTKLib.InputVariable(new int[] { predictionsData.Length }, m_dataType);
var sut = Losses.MeanSquaredError(predictionsVariable, targetsVariable);
var actual = Evaluate(sut, targetsVariable, targetsData,
predictionsVariable, predictionsData);
Assert.AreEqual(0.18f, actual, 0.00001);
}