mandlm
/
Neuro
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Finished the max-value net (2/3/1 neurons) with 10k learning iterations. No good.

main
mandlm 2015-10-22 22:09:35 +02:00
parent d4a22ecae7
commit 6ed30e56c4
3 changed files with 27 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
Layer.cpp
View File

@ -54,16 +54,16 @@ void Layer::updateInputWeights(Layer & prevLayer)
{ {
static const double trainingRate = 0.2; static const double trainingRate = 0.2;
for (size_t currentLayerIndex = 0; currentLayerIndex < sizeWithoutBiasNeuron(); ++currentLayerIndex) for (size_t targetLayerIndex = 0; targetLayerIndex < sizeWithoutBiasNeuron(); ++targetLayerIndex)
{ {
Neuron &targetNeuron = at(currentLayerIndex); const Neuron &targetNeuron = at(targetLayerIndex);
for (size_t prevLayerIndex = 0; prevLayerIndex < prevLayer.size(); ++prevLayerIndex) for (size_t sourceLayerIndex = 0; sourceLayerIndex < prevLayer.size(); ++sourceLayerIndex)
{ {
Neuron &sourceNeuron = prevLayer.at(prevLayerIndex); Neuron &sourceNeuron = prevLayer.at(sourceLayerIndex);
sourceNeuron.setOutputWeight(currentLayerIndex, sourceNeuron.setOutputWeight(targetLayerIndex,
sourceNeuron.getOutputWeight(currentLayerIndex) + sourceNeuron.getOutputWeight(targetLayerIndex) +
sourceNeuron.getOutputValue() * targetNeuron.getGradient() * trainingRate); sourceNeuron.getOutputValue() * targetNeuron.getGradient() * trainingRate);
} }
} }

5
Net.cpp
View File

@ -73,11 +73,10 @@ void Net::backProp(const std::vector<double> &targetValues)
for (unsigned int i = 0; i < numResultValues; ++i) for (unsigned int i = 0; i < numResultValues; ++i)
{ {
double delta = resultValues[i] - targetValues[i]; rmsError += std::pow(resultValues[i] - targetValues[i], 2);
rmsError += delta * delta;
} }
rmsError = sqrt(rmsError / numResultValues); rmsError = std::sqrt(rmsError / numResultValues);
// calculate output neuron gradients // calculate output neuron gradients
for (unsigned int i = 0; i < numResultValues; ++i) for (unsigned int i = 0; i < numResultValues; ++i)

31
Neuro.cpp
View File

@ -12,25 +12,32 @@ int main()
Net myNet({ 2, 3, 1 }); Net myNet({ 2, 3, 1 });
std::vector<double> inputValues = size_t numIterations = 10000;
for (size_t iteration = 0; iteration < numIterations; ++iteration)
{ {
0.1, std::vector<double> inputValues =
0.7, {
}; std::rand() / (double)RAND_MAX,
std::rand() / (double)RAND_MAX
};
std::vector<double> targetValues = { 0.7 }; std::vector<double> targetValues =
{
*std::max_element(inputValues.begin(), inputValues.end())
};
myNet.feedForward(inputValues); myNet.feedForward(inputValues);
std::vector<double> outputValues = myNet.getOutput(); std::vector<double> outputValues = myNet.getOutput();
double error = outputValues[0] - targetValues[0]; double error = outputValues[0] - targetValues[0];
std::cout << "Error: "; std::cout << "Error: ";
std::cout << std::abs(error); std::cout << std::abs(error);
std::cout << std::endl; std::cout << std::endl;
myNet.backProp(targetValues); myNet.backProp(targetValues);
}
} }
catch (std::exception &ex) catch (std::exception &ex)
{ {