private void Test()
{
UnityEngine.Profiling.Profiler.BeginSample("Test");
int correctTestLabels = 0;
for (int i = 0; i < DataManager.Test.NumImgs; i++)
{
int lbl = (int)DataManager.Test.Labels[i];
var handle = DataManager.CopyInput(_net.Inputs, DataManager.Test, i);
handle = NeuralJobs.ForwardPass(_net, handle);
handle.Complete();
int predictedLbl = NeuralMath.ArgMax(_net.Last.Outputs);
if (predictedLbl == lbl)
{
correctTestLabels++;
}
}
float accuracy = correctTestLabels / (float)DataManager.Test.NumImgs;
Debug.Log("Test Accuracy: " + System.Math.Round(accuracy * 100f, 4) + "%");
UnityEngine.Profiling.Profiler.EndSample();
}
public void Execute()
{
for (int i = 0; i < Data.Length; i++)
{
Data[i] = NeuralMath.Sigmoid(Data[i]);
}
}
public void Execute()
{
for (int n = 0; n < Outputs.Length; n++)
{
float dOdZ = NeuralMath.SigmoidPrime(Outputs[n]);
DCDZ[n] = DCDZNext[n] * dOdZ;
}
}
public static void Initialize(FCNetwork net, ref Rng rng)
{
// Todo: init as jobs too. Needs Burst-compatible RNG.
for (int l = 0; l < net.Layers.Length; l++)
{
NeuralMath.RandomGaussian(ref rng, net.Layers[l].Weights, 0f, 1f);
NeuralMath.RandomGaussian(ref rng, net.Layers[l].Biases, 0f, 1f);
}
}
public void Execute()
{
for (int n = 0; n < Outputs.Length; n++)
{
float dOdZ = NeuralMath.SigmoidPrime(Outputs[n]);
float dcdzn = 0f;
for (int nNext = 0; nNext < DCDZNext.Length; nNext++)
{
dcdzn += DCDZNext[nNext] * WeightsNext[nNext * DCDZ.Length + n];
}
DCDZ[n] = dcdzn * dOdZ;
}
}
// Todo: show user all possible integer solutions, given partially filled config
public static int GetOutputWidth(int inWidth, int kernelWidth, int stride, int padding)
{
if (NeuralMath.IsEven(kernelWidth))
{
return(-1);
}
float result = (inWidth - kernelWidth + padding * 2.0f) / (float)stride + 1.0f;
if (result - (int)result < float.Epsilon)
{
return((int)result);
}
return(-1);
}
private void TrainMinibatch()
{
UnityEngine.Profiling.Profiler.BeginSample("TrainMiniBatch");
float avgTrainCost = 0f;
DataManager.GetBatch(_batch, DataManager.Train, ref _rng);
var handle = NeuralJobs.ZeroGradients(_gradientsAvg);
for (int i = 0; i < _batch.Length; i++)
{
handle = DataManager.CopyInput(_net.Inputs, DataManager.Train, _batch[i], handle);
handle = NeuralJobs.ForwardPass(_net, handle);
int lbl = (int)DataManager.Train.Labels[_batch[i]];
handle.Complete();
NeuralMath.ClassToOneHot(lbl, _targetOutputs); // Todo: job
handle = NeuralJobs.BackwardsPass(_net, _gradients, _targetOutputs, handle);
handle = NeuralJobs.AddGradients(_gradients, _gradientsAvg, handle);
handle.Complete();
// Todo: backwards pass logic now does this, don't redo, just check
NeuralMath.Subtract(_targetOutputs, _net.Last.Outputs, _dCdO);
float cost = NeuralMath.Cost(_dCdO);
avgTrainCost += cost;
}
// Update weights and biases according to averaged gradient and learning rate
_rate = 3.0f / (float)BatchSize;
handle = NeuralJobs.UpdateParameters(_net, _gradientsAvg, _rate, handle);
handle.Complete(); // Todo: Is this one needed?
_batchCount++;
avgTrainCost /= (float)BatchSize;
_trainingLoss = (float)System.Math.Round(avgTrainCost, 6);
UnityEngine.Profiling.Profiler.EndSample();
}
public void Execute(int i)
{
Data[i] = NeuralMath.Sigmoid(Data[i]);
}
