public void NN_backpropagation_2L_nb_custom_activation_SigLin_all_training_samples()
{
initData_dataset_naive_bayes_jason_example();
Build2LBackPropagation build =
new Build2LBackPropagation();
build.SetParameters(1, .01, 2000, .01);
build.SetActivationFunction(0, new Dasmic.MLLib.Algorithms.NeuralNetwork.Support.ActivationFunction.Sigmoid());
build.SetActivationFunction(1, new Dasmic.MLLib.Algorithms.NeuralNetwork.Support.ActivationFunction.Linear());
ModelBackPropagationBase model =
(ModelBackPropagationBase)build.BuildModel(
_trainingData, _attributeHeaders,
_indexTargetAttribute);
int count = 0;
for (int row = 0; row < 10; row++)
{
double[] data = GetSingleTrainingRowDataForTest(row);
double value = model.RunModelForSingleData(data);
if (SupportFunctions.DoubleCompare(value,
_trainingData[_indexTargetAttribute][row]))
{
count++;
}
}
Assert.AreEqual(count,
8);
}
public void NN_backpropagation_2L_nb_all_training_samples()
{
initData_dataset_naive_bayes_jason_example();
Build2LBackPropagation build =
new Build2LBackPropagation();
build.SetParameters(1);
ModelBackPropagationBase model =
(ModelBackPropagationBase)build.BuildModel(
_trainingData, _attributeHeaders,
_indexTargetAttribute);
int count = 0;
for (int row = 0; row < 10; row++)
{
double[] data = GetSingleTrainingRowDataForTest(row);
double value = model.RunModelForSingleData(data);
if (SupportFunctions.DoubleCompare(value,
_trainingData[_indexTargetAttribute][row]))
{
count++;
}
}
//Due to random weights
Assert.IsTrue(count >= 8 && count <= 10);
}
public void NN_backpropagation_generic_std_one_hidden_pythagoras_single_data_0()
{
Init_dataset_pythagoras();
BuildGenericBackPropagationStandard build =
new BuildGenericBackPropagationStandard();
build.SetParameters(0, 1, .001, 2500, .001);
//build.SetNumberOfHiddenLayers(1);
build.AddHiddenLayer(0, 2, new Sigmoid());
build.SetOutputLayerActivationFunction(new Linear());
ModelBackPropagationBase model =
(ModelBackPropagationBase)build.BuildModel(
_trainingData, _attributeHeaders,
_indexTargetAttribute);
/*
* model.SetWeight(1, 0, 0, .2196);
* model.SetWeight(1, 1, 0, .121);
* model.SetWeight(1, 2, 0, -4.18);
*
* model.SetWeight(1, 0, 1, .15367);
* model.SetWeight(1, 1, 1, .2216);
* model.SetWeight(1, 2, 1, -.99404);
*
* model.SetWeight(2, 0, 0, 15.43);
* model.SetWeight(2, 1, 0, 12.92);
* model.SetWeight(2, 2, 0, -3.33);*/
System.Diagnostics.Debug.WriteLine("Weight[1][0][0]:" + model.GetWeight(1, 0, 0));
System.Diagnostics.Debug.WriteLine("Weight[1][1][0]:" + model.GetWeight(1, 1, 0));
System.Diagnostics.Debug.WriteLine("Weight[1][2][0]:" + model.GetWeight(1, 2, 0));
System.Diagnostics.Debug.WriteLine("Weight[1][0][1]:" + model.GetWeight(1, 0, 1));
System.Diagnostics.Debug.WriteLine("Weight[1][1][1]:" + model.GetWeight(1, 1, 1));
System.Diagnostics.Debug.WriteLine("Weight[1][2][1]:" + model.GetWeight(1, 2, 1));
System.Diagnostics.Debug.WriteLine("Weight[2][0][1]:" + model.GetWeight(2, 0, 0));
System.Diagnostics.Debug.WriteLine("Weight[2][1][1]:" + model.GetWeight(2, 1, 0));
System.Diagnostics.Debug.WriteLine("Weight[2][2][1]:" + model.GetWeight(2, 2, 0));
int row = 0;
double[] data = GetSingleTrainingRowDataForTest(row);
double value = model.RunModelForSingleData(data);
System.Diagnostics.Debug.WriteLine("Final value:" + value);
//Actual answer is 1.41
Assert.IsTrue(value > 1.0 && value < 3.0);
}
public void NN_backpropagation_generic_rprop_one_hidden_pythagoras_rmse_data_1()
{
Init_dataset_pythagoras();
BuildGenericBackPropagationRprop build =
new BuildGenericBackPropagationRprop();
build.SetParameters(0, 1, .02, 20000, .1);//,.005,2000);
build.AddHiddenLayer(0, 2, new Sigmoid());
build.SetOutputLayerActivationFunction(new Linear());
ModelBackPropagationBase model =
(ModelBackPropagationBase)build.BuildModel(
_trainingData, _attributeHeaders,
_indexTargetAttribute);
/*
* model.SetWeight(1, 0, 0, .53);
* model.SetWeight(1, 1, 0, .53);
* model.SetWeight(1, 2, 0, .53);
*
* model.SetWeight(1, 0, 1, .53);
* model.SetWeight(1, 1, 1, .53);
* model.SetWeight(1, 2, 1, .53);
*
* model.SetWeight(2, 0, 0, .53);
* model.SetWeight(2, 1, 0, .53);
* model.SetWeight(2, 2, 0, .53);
*/
System.Diagnostics.Debug.WriteLine("Weight[1][0][0]:" + model.GetWeight(1, 0, 0));
System.Diagnostics.Debug.WriteLine("Weight[1][1][0]:" + model.GetWeight(1, 1, 0));
System.Diagnostics.Debug.WriteLine("Weight[1][2][0]:" + model.GetWeight(1, 2, 0));
System.Diagnostics.Debug.WriteLine("Weight[1][0][1]:" + model.GetWeight(1, 0, 1));
System.Diagnostics.Debug.WriteLine("Weight[1][1][1]:" + model.GetWeight(1, 1, 1));
System.Diagnostics.Debug.WriteLine("Weight[1][2][1]:" + model.GetWeight(1, 2, 1));
System.Diagnostics.Debug.WriteLine("Weight[2][0][1]:" + model.GetWeight(2, 0, 0));
System.Diagnostics.Debug.WriteLine("Weight[2][1][1]:" + model.GetWeight(2, 1, 0));
System.Diagnostics.Debug.WriteLine("Weight[2][2][1]:" + model.GetWeight(2, 2, 0));
int row = 0;
double[] data = GetSingleTrainingRowDataForTest(row);
double value = model.RunModelForSingleData(data);
Assert.IsTrue(value <= 1.69);
}
public void NN_backpropagation_generic_rprop_missing_layer_throws_exception()
{
initData_dataset_gaussian_naive_bayes_jason_example();
BuildGenericBackPropagationRprop build =
new BuildGenericBackPropagationRprop();
build.SetParameters(1, 3, .5, 1500);
build.AddHiddenLayer(0, 2, new Sigmoid());
build.AddHiddenLayer(1, 2, new Sigmoid());
build.SetOutputLayerActivationFunction(new Sigmoid());
ModelBackPropagationBase model =
(ModelBackPropagationBase)build.BuildModel(
_trainingData, _attributeHeaders,
_indexTargetAttribute);
}
protected bool VerifyLayers(ModelBackPropagationBase model)
{
bool exceptionFlag = false;
Parallel.For(0, model.GetNumberOfLayers() - 1,
new ParallelOptions {
MaxDegreeOfParallelism = _maxParallelThreads
},
idxLayer =>
{
if (model.GetLayer(idxLayer) == null)
{
exceptionFlag = true;
}
});
return(exceptionFlag);
}
public void Deep_NN_generic_jason_simple_rmse()
{
Init_dataset_jason_linear_regression();
BuildGenericDeepNN build =
new BuildGenericDeepNN();
build.SetParameters(0, 2);
//Use Default Parameters
ModelBackPropagationBase model =
(ModelBackPropagationBase)build.BuildModel(
_trainingData, _attributeHeaders,
_indexTargetAttribute);
double value = model.GetModelRMSE(_trainingData);
Assert.IsTrue(value < .61 && value > 0);
}
public void NN_backpropagation_generic_rprop_no_hidden_pythagoras_rmse()
{
Init_dataset_pythagoras();
BuildGenericBackPropagationRprop build =
new BuildGenericBackPropagationRprop();
build.SetParameters(0, 0, .3, 1);
build.SetOutputLayerActivationFunction(new Linear());
ModelBackPropagationBase model =
(ModelBackPropagationBase)build.BuildModel(
_trainingData, _attributeHeaders,
_indexTargetAttribute);
double value = model.GetModelRMSE(_trainingData);
Assert.IsTrue(value < 10);
}
public void NN_backpropagation_generic_rprop_jason_simple_rmse()
{
Init_dataset_jason_linear_regression();
BuildGenericBackPropagationRprop build =
new BuildGenericBackPropagationRprop();
build.SetParameters(0, 1, .01, 10000);
build.AddHiddenLayer(0, 2, new Sigmoid());
build.SetOutputLayerActivationFunction(new Linear());
ModelBackPropagationBase model =
(ModelBackPropagationBase)build.BuildModel(
_trainingData, _attributeHeaders,
_indexTargetAttribute);
double value = model.GetModelRMSE(_trainingData);
Assert.IsTrue(value < .61 && value > 0);
}
public void NN_backpropagation_2L_jason_simple_rmse()
{
Init_dataset_jason_linear_regression();
Build2LBackPropagation build =
new Build2LBackPropagation();
build.SetParameters(0, .01, 3000, .05);
build.SetActivationFunction(0, new Dasmic.MLLib.Algorithms.NeuralNetwork.Support.ActivationFunction.Sigmoid());
build.SetActivationFunction(1, new Dasmic.MLLib.Algorithms.NeuralNetwork.Support.ActivationFunction.Linear());
ModelBackPropagationBase model =
(ModelBackPropagationBase)build.BuildModel(
_trainingData, _attributeHeaders,
_indexTargetAttribute);
double value = model.GetModelRMSE(_trainingData);
Assert.IsTrue(value < 1.0);
}
public void NN_backpropagation_generic_std_no_hidden_pythagoras_rmse()
{
Init_dataset_pythagoras();
BuildGenericBackPropagationStandard build =
new BuildGenericBackPropagationStandard();
build.SetParameters(0, 0, .3, 1);
//build.SetNumberOfHiddenLayers(0);
//build.AddHiddenLayer(0, 2, new Sigmoid());
build.SetOutputLayerActivationFunction(new Linear());
ModelBackPropagationBase model =
(ModelBackPropagationBase)build.BuildModel(
_trainingData, _attributeHeaders,
_indexTargetAttribute);
double value = model.GetModelRMSE(_trainingData);
Assert.IsTrue(value > 100000);
}
public void NN_backpropagation_2L_gnb_single_training_sample_class_0()
{
initData_dataset_gaussian_naive_bayes_jason_example();
Build2LBackPropagation build =
new Build2LBackPropagation();
build.SetParameters(1);
ModelBackPropagationBase model =
(ModelBackPropagationBase)build.BuildModel(
_trainingData, _attributeHeaders,
_indexTargetAttribute);
int row = 0;
double[] data = GetSingleTrainingRowDataForTest(row);
double value = model.RunModelForSingleData(data);
Assert.AreEqual(value,
_trainingData[_indexTargetAttribute][row]);
}
public void NN_backpropagation_generic_rprop_gnb_single_training_sample_class_0()
{
initData_dataset_gaussian_naive_bayes_jason_example();
BuildGenericBackPropagationRprop build =
new BuildGenericBackPropagationRprop();
build.SetParameters(1, 1);
build.AddHiddenLayer(0, 2, new Sigmoid());
build.SetOutputLayerActivationFunction(new Sigmoid());
ModelBackPropagationBase model =
(ModelBackPropagationBase )build.BuildModel(
_trainingData, _attributeHeaders,
_indexTargetAttribute);
int row = 0;
double[] data = GetSingleTrainingRowDataForTest(row);
double value = model.RunModelForSingleData(data);
Assert.AreEqual(value,
_trainingData[_indexTargetAttribute][row]);
}
