public EncogTANHActivation()
: base()
{
activationFunction = new ActivationTANH();
activationLevel = 1.0;
activationMin = -1.0;
activationMax = 1.0;
}
public void TestTANH()
{
var activation = new ActivationTANH();
Assert.IsTrue(activation.HasDerivative);
var clone = (ActivationTANH) activation.Clone();
Assert.IsNotNull(clone);
double[] input = {0.0};
activation.ActivationFunction(input, 0, 1);
Assert.AreEqual(0.0, input[0], 0.1);
// test derivative, should throw an error
input[0] = activation.DerivativeFunction(input[0],input[0]);
Assert.AreEqual(1.0, input[0], 0.1);
}
public void TestTANH()
{
var activation = new ActivationTANH();
Assert.IsTrue(activation.HasDerivative);
var clone = activation.Clone();
Assert.IsInstanceOfType(clone, typeof(ActivationTANH));
double[] input = { 0.0 };
activation.ActivationFunction(input, 0, 1);
Assert.AreEqual(0.0, input[0], 0.1);
input[0] = activation.DerivativeFunction(input[0], input[0]);
Assert.AreEqual(1.0, input[0], 0.1);
}
public void TestTANH()
{
var activation = new ActivationTANH();
Assert.IsTrue(activation.HasDerivative);
var clone = (ActivationTANH)activation.Clone();
Assert.IsNotNull(clone);
double[] input = { 0.0 };
activation.ActivationFunction(input, 0, 1);
Assert.AreEqual(0.0, input[0], 0.1);
// test derivative, should throw an error
input[0] = activation.DerivativeFunction(input[0], input[0]);
Assert.AreEqual(1.0, input[0], 0.1);
}
public FlatNetwork(int input, int hidden1, int hidden2, int output, bool tanh)
{
// This item is obfuscated and can not be translated.
FlatLayer[] layerArray;
IActivationFunction function2;
int num;
IActivationFunction function3;
IActivationFunction activation = new ActivationLinear();
if (tanh)
{
goto Label_01EA;
}
goto Label_0219;
Label_000B:
this._connectionLimit = 0.0;
this.Init(layerArray);
if (0 != 0)
{
if (((uint) hidden1) > uint.MaxValue)
{
goto Label_0069;
}
goto Label_0219;
}
if (((uint) num) < 0)
{
goto Label_0189;
}
return;
Label_0069:
layerArray[3] = new FlatLayer(function2, output, 0.0);
if ((((uint) num) + ((uint) output)) >= 0)
{
}
Label_0098:
this._isLimited = false;
goto Label_000B;
Label_00A1:
if (hidden2 != 0)
{
if ((((uint) tanh) - ((uint) hidden2)) > uint.MaxValue)
{
goto Label_01AA;
}
layerArray = new FlatLayer[4];
layerArray[0] = new FlatLayer(activation, input, 1.0);
layerArray[1] = new FlatLayer(function2, hidden1, 1.0);
if ((((uint) tanh) + ((uint) hidden2)) <= uint.MaxValue)
{
layerArray[2] = new FlatLayer(function2, hidden2, 1.0);
if (((uint) hidden2) > uint.MaxValue)
{
goto Label_015F;
}
goto Label_0069;
}
goto Label_000B;
}
Label_010F:
num = Math.Max(hidden1, hidden2);
layerArray = new FlatLayer[] { new FlatLayer(activation, input, 1.0), new FlatLayer(function2, num, 1.0), new FlatLayer(function2, output, 0.0) };
goto Label_0098;
Label_015F:
if (((uint) output) > uint.MaxValue)
{
goto Label_01AA;
}
if ((((uint) tanh) + ((uint) num)) <= uint.MaxValue)
{
goto Label_0198;
}
Label_0189:
if (-1 != 0)
{
goto Label_00A1;
}
goto Label_010F;
Label_0198:
if (hidden1 != 0)
{
goto Label_00A1;
}
goto Label_010F;
Label_01AA:
if (hidden2 != 0)
{
goto Label_0198;
}
layerArray = new FlatLayer[] { new FlatLayer(activation, input, 1.0), new FlatLayer(function2, output, 0.0) };
goto Label_0098;
Label_01EA:
function2 = new ActivationTANH();
if (hidden1 != 0)
{
goto Label_015F;
}
goto Label_01AA;
Label_0219:
function3 = new ActivationSigmoid();
if ((((uint) num) | 2) != 0)
{
goto Label_01EA;
}
goto Label_010F;
}
/// <summary>
/// Initializes the neural network prior to the training. The actual activation output
/// for the output layer and its maximum and minimum levels are also defined here.
/// </summary>
/// <returns></returns>
private BasicNetwork PrepareNetwork()
{
IActivationFunction outputLayerActivation;
EncogActivation encogActivation = (EncogActivation)base.activationFunction;
_activationMin = base.activationFunction.activationMin;
_activationMax = base.activationFunction.activationMax;
// Checking if the selected activation function is adequate for the
// output layer and changing it if needed. The activation function in the
// last neuron layer should produce outputs between [0,1] or [-1,1].
if (_activationMin < -1 && _activationMax > 1)
{
//We will use a TANH activation function for the output layer
outputLayerActivation = new ActivationTANH();
_activationMin = -1;
_activationMax = 1;
}
else if (_activationMin == 0 && _activationMax > 1)
{
//We will use a Sigmoid activation function for the output layer
outputLayerActivation = new ActivationSigmoid();
_activationMin = 0;
_activationMax = 1;
}
else
{
outputLayerActivation = ((EncogActivation)base.activationFunction).CreateInstance();
}
BasicNetwork network = new BasicNetwork();
if (autoHiddenLayerSize)
_actualHiddenLayerSize = inputDim * hiddenLayerSizeFactor;
else _actualHiddenLayerSize = hiddenLayerSize;
network.AddLayer(new BasicLayer(null, true, inputDim));
network.AddLayer(new BasicLayer(encogActivation.CreateInstance(), true, _actualHiddenLayerSize));
network.AddLayer(new BasicLayer(outputLayerActivation, false, outputDim));
network.Structure.FinalizeStructure();
network.Reset();
return network;
}
void AddLayers(List<LayerConfig> gen)
{
foreach (var g in gen)
{
IActivationFunction act;
if (g.ActivationType == 0)
{
act = new ActivationBiPolar();
}
switch (g.ActivationType )
{
case 0:
act = new ActivationBiPolar();
break;
case 1:
act = new ActivationBipolarSteepenedSigmoid ();
break;
case 2:
act = new ActivationClippedLinear();
break;
case 3:
act = new ActivationCompetitive();
break;
case 4:
act = new ActivationElliott();
break;
case 5:
act = new ActivationElliottSymmetric();
break;
case 6:
act = new ActivationGaussian();
break;
case 7:
act = new ActivationLinear();
break;
case 8:
act = new ActivationLOG();
break;
case 9:
act = new ActivationRamp();
break;
case 10:
act = new ActivationRamp();
break;
case 11:
act = new ActivationSigmoid();
break;
case 12:
act = new ActivationSIN();
break;
case 13:
act = new ActivationSoftMax();
break;
case 14:
act = new ActivationSteepenedSigmoid();
break;
case 15:
act = new ActivationStep();
break;
case 16:
act = new ActivationTANH();
break;
default:
act = new ActivationSoftMax();
break;
}
network.AddLayer(new BasicLayer(act, g.hasBias, g.neurons));
}
}