} // Main
static double[][] MakeData(int numItems, int numInput, int[] numHidden, int numOutput, int seed)
{
// generate data using a Deep NN (tanh hidden activation)
DeepNet dn = new DeepNet(numInput, numHidden, numOutput); // make a DNN generator
Random rrnd = new Random(seed); // to make random weights & biases, random input vals
double wtLo = -9.0;
double wtHi = 9.0;
int nw = DeepNet.NumWeights(numInput, numHidden, numOutput);
double[] wts = new double[nw];
for (int i = 0; i < nw; ++i)
{
wts[i] = (wtHi - wtLo) * rrnd.NextDouble() + wtLo;
}
dn.SetWeights(wts);
double[][] result = new double[numItems][]; // make the result matrix holder
for (int r = 0; r < numItems; ++r)
{
result[r] = new double[numInput + numOutput]; // allocate the cols
}
double inLo = -4.0; // pseudo-Gaussian scaling
double inHi = 4.0;
for (int r = 0; r < numItems; ++r) // each row
{
double[] inputs = new double[numInput]; // random input values
for (int i = 0; i < numInput; ++i)
{
inputs[i] = (inHi - inLo) * rrnd.NextDouble() + inLo;
}
//ShowVector(inputs, 2);
double[] probs = dn.ComputeOutputs(inputs); // compute the outputs (as softmax probs) like [0.10, 0.15, 0.55, 0.20]
//dn.Dump();
//Console.ReadLine();
//ShowVector(probs, 4);
double[] outputs = ProbsToClasses(probs); // convert to outputs like [0, 0, 1, 0]
int c = 0;
for (int i = 0; i < numInput; ++i)
{
result[r][c++] = inputs[i];
}
for (int i = 0; i < numOutput; ++i)
{
result[r][c++] = outputs[i];
}
//Console.WriteLine("");
}
return(result);
} // MakeData
