public void CanFeedForwardCorrectlySingleInputNeuronSingleHiddenLayerWithSingleNeuronMultipleOutputNeurons()
{
FeedForwardTester()
.NeuralNetworkEnvironment(TestContext, PredictableGenerator)
.TargetNeuralNetwork(nn => nn
.InputLayer(l => l
.Neurons(
n => n.Id(1).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid)))
.HiddenLayer(l => l
.Neurons(
n => n.Id(2).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid)))
.OutputLayer(l => l
.Neurons(
n => n.Id(3).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(4).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid)))
.Synapses(
s => s.SynapseBetween(inputNeuronId: 1, outputNeuronId: 2, weight: 0.9812),
s => s.SynapseBetween(inputNeuronId: 2, outputNeuronId: 3, weight: 0.1),
s => s.SynapseBetween(inputNeuronId: 2, outputNeuronId: 4, weight: 0.6209)))
.InputAndExpectOutput(
inputs: new [] { 0.000102 },
expectedOutput: new [] { 0.5125, 0.577 },
parallelOptions: ParallelOptionsExtensions.SingleThreadedOptions)
.InputAndExpectOutput(
inputs: new [] { 0.000102 },
expectedOutput: new [] { 0.5125, 0.577 },
parallelOptions: ParallelOptionsExtensions.MultiThreadedOptions(50));
}
public async Task TrainConfiguredNetworkForEpochs(
int epochs,
NeuralNetworkTrainingConfiguration trainingConfig)
{
var trainingData = dataProvider.TrainingData
.Select(transaction => transaction.ToTrainingData())
.ToList();
await TrainingController.For(BackPropagation.WithConfiguration(
networkAccessor.TargetNetwork,
ParallelOptionsExtensions.MultiThreadedOptions(trainingConfig.ThreadCount),
trainingConfig.LearningRate,
trainingConfig.Momentum))
.TrainForEpochsOrErrorThresholdMet(trainingData, epochs, trainingConfig.MinimumErrorThreshold);
}
public void CanFeedForwardCorrectlyMultipleInputNeuronsMultipleHiddenLayersMultipleOutputNeurons()
{
FeedForwardTester()
.NeuralNetworkEnvironment(TestContext, PredictableGenerator)
.TargetNeuralNetwork(nn => nn
.InputLayer(l => l
.Neurons(
n => n.Id(1).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(2).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid)))
.HiddenLayer(l => l
.Neurons(
n => n.Id(3).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(4).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid)))
.HiddenLayer(l => l
.Neurons(
n => n.Id(5).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid)))
.HiddenLayer(l => l
.Neurons(
n => n.Id(6).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid)))
.OutputLayer(l => l
.Neurons(
n => n.Id(7).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(8).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(9).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid)))
.Synapses(
s => s.SynapseBetween(inputNeuronId: 1, outputNeuronId: 3, weight: 0.67),
s => s.SynapseBetween(inputNeuronId: 1, outputNeuronId: 4, weight: 0.72),
s => s.SynapseBetween(inputNeuronId: 2, outputNeuronId: 3, weight: 0.145),
s => s.SynapseBetween(inputNeuronId: 2, outputNeuronId: 4, weight: 0.134),
s => s.SynapseBetween(inputNeuronId: 3, outputNeuronId: 5, weight: 0.9),
s => s.SynapseBetween(inputNeuronId: 4, outputNeuronId: 5, weight: 0.134),
s => s.SynapseBetween(inputNeuronId: 5, outputNeuronId: 6, weight: 0.11),
s => s.SynapseBetween(inputNeuronId: 6, outputNeuronId: 7, weight: 0.781),
s => s.SynapseBetween(inputNeuronId: 6, outputNeuronId: 8, weight: 0.7101),
s => s.SynapseBetween(inputNeuronId: 6, outputNeuronId: 9, weight: 0.7101)))
.InputAndExpectOutput(
inputs: new [] { 0.9145, 0.7812 },
expectedOutput: new [] { 0.59985, 0.591, 0.591 },
parallelOptions: ParallelOptionsExtensions.SingleThreadedOptions)
.InputAndExpectOutput(
inputs: new [] { 0.9145, 0.7812 },
expectedOutput: new [] { 0.59985, 0.591, 0.591 },
parallelOptions: ParallelOptionsExtensions.MultiThreadedOptions(10));
}
public void CanFeedForwardOutputCorrectlyForMultipleInputNeuronsNoHiddenLayersSingleOutputNeuron()
{
FeedForwardTester()
.NeuralNetworkEnvironment(TestContext, PredictableGenerator)
.TargetNeuralNetwork(nn => nn
.InputLayer(l => l
.Neurons(
n => n.Id(1).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(2).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid)))
.OutputLayer(l => l
.Neurons(n => n.Id(3).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid)))
.Synapses(
s => s.SynapseBetween(inputNeuronId: 1, outputNeuronId: 3, weight: 0.87),
s => s.SynapseBetween(inputNeuronId: 2, outputNeuronId: 3, weight: 0.11)))
.InputAndExpectOutput(
inputs: new [] { 0.4, 0.56 },
expectedOutput: new [] { 0.60099 },
parallelOptions: ParallelOptionsExtensions.SingleThreadedOptions)
.InputAndExpectOutput(
inputs: new [] { 0.4, 0.56 },
expectedOutput: new [] { 0.60099 },
parallelOptions: ParallelOptionsExtensions.MultiThreadedOptions(15));
}
public void CanFeedForwardCorrectlyInComplexMultiLayerNeuralNetwork()
{
FeedForwardTester()
.NeuralNetworkEnvironment(TestContext, PredictableGenerator)
.TargetNeuralNetwork(nn => nn
.InputLayer(l => l
.Neurons(
n => n.Id(1).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(2).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(3).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(4).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(5).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid)))
.HiddenLayer(l => l
.Neurons(
n => n.Id(6).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(7).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(8).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(9).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid)))
.HiddenLayer(l => l
.Neurons(
n => n.Id(10).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(11).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(12).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid)))
.HiddenLayer(l => l
.Neurons(
n => n.Id(13).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(14).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(15).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(16).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid)))
.OutputLayer(l => l
.Neurons(
n => n.Id(17).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(18).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(19).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid),
n => n.Id(20).ErrorGradient(0).Output(0).Activation(ActivationType.Sigmoid)))
.Synapses(
s => s.SynapseBetween(inputNeuronId: 1, outputNeuronId: 6, weight: 0.67),
s => s.SynapseBetween(inputNeuronId: 1, outputNeuronId: 7, weight: 0.72),
s => s.SynapseBetween(inputNeuronId: 1, outputNeuronId: 8, weight: 0.145),
s => s.SynapseBetween(inputNeuronId: 1, outputNeuronId: 9, weight: 0.134),
s => s.SynapseBetween(inputNeuronId: 2, outputNeuronId: 6, weight: 0.999),
s => s.SynapseBetween(inputNeuronId: 2, outputNeuronId: 7, weight: 0.13),
s => s.SynapseBetween(inputNeuronId: 2, outputNeuronId: 8, weight: 0.2874),
s => s.SynapseBetween(inputNeuronId: 2, outputNeuronId: 9, weight: 0.9),
s => s.SynapseBetween(inputNeuronId: 3, outputNeuronId: 6, weight: 0.7823),
s => s.SynapseBetween(inputNeuronId: 3, outputNeuronId: 7, weight: 0.234),
s => s.SynapseBetween(inputNeuronId: 3, outputNeuronId: 8, weight: 0.981),
s => s.SynapseBetween(inputNeuronId: 3, outputNeuronId: 9, weight: 0.99999),
s => s.SynapseBetween(inputNeuronId: 4, outputNeuronId: 6, weight: 0.11),
s => s.SynapseBetween(inputNeuronId: 4, outputNeuronId: 7, weight: 0.2687),
s => s.SynapseBetween(inputNeuronId: 4, outputNeuronId: 8, weight: 0.6),
s => s.SynapseBetween(inputNeuronId: 4, outputNeuronId: 9, weight: 0.614),
s => s.SynapseBetween(inputNeuronId: 5, outputNeuronId: 6, weight: 0.781),
s => s.SynapseBetween(inputNeuronId: 5, outputNeuronId: 7, weight: 0.7101),
s => s.SynapseBetween(inputNeuronId: 5, outputNeuronId: 8, weight: 0.9898),
s => s.SynapseBetween(inputNeuronId: 5, outputNeuronId: 9, weight: 0.3333),
s => s.SynapseBetween(inputNeuronId: 6, outputNeuronId: 10, weight: 0.1256),
s => s.SynapseBetween(inputNeuronId: 6, outputNeuronId: 11, weight: 0.876),
s => s.SynapseBetween(inputNeuronId: 6, outputNeuronId: 12, weight: 0.9999),
s => s.SynapseBetween(inputNeuronId: 7, outputNeuronId: 10, weight: 0.11333),
s => s.SynapseBetween(inputNeuronId: 7, outputNeuronId: 11, weight: 0.5555),
s => s.SynapseBetween(inputNeuronId: 7, outputNeuronId: 12, weight: 0.187),
s => s.SynapseBetween(inputNeuronId: 8, outputNeuronId: 10, weight: 0.7101),
s => s.SynapseBetween(inputNeuronId: 8, outputNeuronId: 11, weight: 0.68463),
s => s.SynapseBetween(inputNeuronId: 8, outputNeuronId: 12, weight: 0.123),
s => s.SynapseBetween(inputNeuronId: 9, outputNeuronId: 10, weight: 0.23412),
s => s.SynapseBetween(inputNeuronId: 9, outputNeuronId: 11, weight: 0.1),
s => s.SynapseBetween(inputNeuronId: 9, outputNeuronId: 12, weight: 0.2333),
s => s.SynapseBetween(inputNeuronId: 10, outputNeuronId: 13, weight: 0.9123),
s => s.SynapseBetween(inputNeuronId: 10, outputNeuronId: 14, weight: 0.9),
s => s.SynapseBetween(inputNeuronId: 10, outputNeuronId: 15, weight: 0.876),
s => s.SynapseBetween(inputNeuronId: 10, outputNeuronId: 16, weight: 0.7),
s => s.SynapseBetween(inputNeuronId: 11, outputNeuronId: 13, weight: 0.7101),
s => s.SynapseBetween(inputNeuronId: 11, outputNeuronId: 14, weight: 0.12457),
s => s.SynapseBetween(inputNeuronId: 11, outputNeuronId: 15, weight: 0.51),
s => s.SynapseBetween(inputNeuronId: 11, outputNeuronId: 16, weight: 0.4999),
s => s.SynapseBetween(inputNeuronId: 12, outputNeuronId: 13, weight: 0.3467),
s => s.SynapseBetween(inputNeuronId: 12, outputNeuronId: 14, weight: 0.9898),
s => s.SynapseBetween(inputNeuronId: 12, outputNeuronId: 15, weight: 0.1956),
s => s.SynapseBetween(inputNeuronId: 12, outputNeuronId: 16, weight: 0.9812),
s => s.SynapseBetween(inputNeuronId: 13, outputNeuronId: 17, weight: 0.2222),
s => s.SynapseBetween(inputNeuronId: 13, outputNeuronId: 18, weight: 0.2),
s => s.SynapseBetween(inputNeuronId: 13, outputNeuronId: 19, weight: 0.231),
s => s.SynapseBetween(inputNeuronId: 13, outputNeuronId: 20, weight: 0.7101),
s => s.SynapseBetween(inputNeuronId: 14, outputNeuronId: 17, weight: 0.4198),
s => s.SynapseBetween(inputNeuronId: 14, outputNeuronId: 18, weight: 0.7199),
s => s.SynapseBetween(inputNeuronId: 14, outputNeuronId: 19, weight: 0.1999),
s => s.SynapseBetween(inputNeuronId: 14, outputNeuronId: 20, weight: 0.5612),
s => s.SynapseBetween(inputNeuronId: 15, outputNeuronId: 17, weight: 0.78561),
s => s.SynapseBetween(inputNeuronId: 15, outputNeuronId: 18, weight: 0.1234),
s => s.SynapseBetween(inputNeuronId: 15, outputNeuronId: 19, weight: 0.6512),
s => s.SynapseBetween(inputNeuronId: 15, outputNeuronId: 20, weight: 0.9823),
s => s.SynapseBetween(inputNeuronId: 16, outputNeuronId: 17, weight: 0.9812),
s => s.SynapseBetween(inputNeuronId: 16, outputNeuronId: 18, weight: 0.1111),
s => s.SynapseBetween(inputNeuronId: 16, outputNeuronId: 19, weight: 0.3333),
s => s.SynapseBetween(inputNeuronId: 16, outputNeuronId: 20, weight: 0.7777)))
.InputAndExpectOutput(
inputs: new [] { 0.9912, 0.81234, 0.0001, 0.45231, 0.67128 },
expectedOutput: new [] { 0.87728, 0.72048, 0.75779, 0.92166 },
parallelOptions: ParallelOptionsExtensions.SingleThreadedOptions)
.InputAndExpectOutput(
inputs: new [] { 0.9912, 0.81234, 0.0001, 0.45231, 0.67128 },
expectedOutput: new [] { 0.87728, 0.72048, 0.75779, 0.92166 },
parallelOptions: ParallelOptionsExtensions.MultiThreadedOptions(10));
}