/// <summary>
/// Greather mults means greather variance from 0 after activation function
/// </summary>
/// <param name="mult1">Optimals are: 1.5 for inputs close to 1 and 5.0 for 10% ones and rest zeroes</param>
/// <param name="mult2">Optimal is 3.5 for all sets of inputs if prev layers was good</param>
public static void RandomizeUniform(ISimpleMLP nn, double mult1 = 5, double mult2 = 3.5)
{
double limit = Sqrt(3.0 / nn.NInput) * mult1;
for (int i = 0; i <= nn.NInput; i++)
{
for (int j = 0; j < nn.NHidden; j++)
{
//set weights to random values
nn.GwInputHidden[j, i] = (rand.NextDouble() * 2 - 1) * limit;
}
}
limit = Sqrt(3.0 / nn.NHidden) * mult2;
for (int i = 0; i <= nn.NHidden; i++)
{
for (int j = 0; j < nn.NOutput; j++)
{
//set weights to random values
nn.GwHiddenOutput[j, i] = (rand.NextDouble() * 2 - 1) * limit;
}
}
}
public static void RandomizeWeights(ISimpleMLP nn)
{
//set weights between input and hidden
//--------------------------------------------------------------------------------------------------------
for (int i = 0; i <= nn.NInput; i++)
{
for (int j = 0; j < nn.NHidden; j++)
{
//set weights to random values
nn.GwInputHidden[j, i] = (rand.NextDouble() * 12 - 6) / nn.NInput;
}
}
//set weights between input and hidden
//--------------------------------------------------------------------------------------------------------
for (int i = 0; i <= nn.NHidden; i++)
{
for (int j = 0; j < nn.NOutput; j++)
{
//set weights to random values
nn.GwHiddenOutput[j, i] = (rand.NextDouble() * 12 - 6) / nn.NHidden;
}
}
}
