/// <summary>
/// Constructs a new Gaussian Weight initialization.
/// </summary>
///
/// <param name="network">The activation network whose weights will be initialized.</param>
/// <param name="stdDev">The standard deviation to be used. Common values lie in the 0.001-
/// 0.1 range. Default is 0.1.</param>
///
public GaussianWeights(ActivationNetwork network, double stdDev = 0.1)
{
this.network = network;
this.random = new GaussianGenerator(0f, (float)stdDev, Accord.Math.Tools.Random.Next());
this.UpdateThresholds = false;
}
/// <summary>
/// Constructs a new Gaussian Weight initialization.
/// </summary>
///
/// <param name="network">The activation network whose weights will be initialized.</param>
/// <param name="stdDev">The standard deviation to be used. Common values lie in the 0.001-
/// 0.1 range. Default is 0.1.</param>
///
public GaussianWeights(ActivationNetwork network, double stdDev = 0.1)
{
this.network = network;
this.random = new GaussianGenerator(0f, (float)stdDev, Accord.Math.Tools.Random.Next());
this.UpdateThresholds = false;
}
//scale bicubic
private static Bitmap Scale(Bitmap image)
{
IRandomNumberGenerator generator = new GaussianGenerator(1.05f, 0.1f);
ResizeBicubic filter = new ResizeBicubic((int)(image.Width * 1.035), (int)(image.Height * 1.035));
// apply the filter
return(filter.Apply(image));
}
//apply couds texture to smudge the image
private static Bitmap Texturize(Bitmap image)
{
IRandomNumberGenerator generator = new GaussianGenerator(-.25f, 0.3f);
// create filter
Texturer filter = new Texturer(new CloudsTexture(), generator.Next(), .995);//.98
// apply the filter
return(filter.Apply(image));
}
//cause gaussian noise spikes
private static Bitmap Speckle(Bitmap image)
{
IRandomNumberGenerator generator = new GaussianGenerator(0f, 5f);
// create filter
AdditiveNoise filter = new AdditiveNoise(generator);
// apply the filter
return(filter.Apply(image));
}
public static void TestGaussian()
{
float t = math.sqrt(-2.0f * math.log(UnityEngine.Random.value)) * math.cos(UnityEngine.Random.value);
Debug.Log(t);
Dictionary <int, int> _cnt1 = new Dictionary <int, int>();
Dictionary <int, int> _cnt2 = new Dictionary <int, int>();
GaussianGenerator rndg = new GaussianGenerator(new Random((uint)UnityEngine.Random.Range(0, int.MaxValue)));
for (int i = 0; i < 100000; i++)
{
float2 r = rndg.NextFloat2();
int2 keys = new int2(Mathf.FloorToInt(r[0] * 20), Mathf.FloorToInt(r[1] * 20));
if (!_cnt1.ContainsKey(keys[0]))
{
_cnt1[keys[0]] = 0;
}
_cnt1[keys[0]] += 1;
if (!_cnt2.ContainsKey(keys[1]))
{
_cnt2[keys[1]] = 0;
}
_cnt2[keys[1]] += 1;
}
int width = math.max(_cnt1.Max(kv => kv.Key), _cnt2.Max(kv => kv.Key));
int height = math.max(_cnt1.Max(kv => kv.Value), _cnt2.Max(kv => kv.Value)) / 10;
for (int i = -width; i < width; i++)
{
{
float val1 = 0.5f / height;
if (_cnt1.ContainsKey(i))
{
val1 = _cnt1[i] / (float)height;
}
Debug.DrawRay(Vector3.right * i / (width / 5f), Vector3.up * val1, Color.blue, 10, false);
}
{
float val2 = 0.5f / height;
if (_cnt2.ContainsKey(i))
{
val2 = _cnt2[i] / (float)height;
}
Debug.DrawRay(Vector3.right * i / (width / 5f) + Vector3.forward / 10, Vector3.up * val2, Color.yellow, 10, false);
}
}
}
private IntVector2 setLevelSize(int seed)
{
System.Random rand = new System.Random(seed);
double mean = (_levelSettings.minLevelSize + _levelSettings.maxLevelSize) / 2;
float sizeX = (float)GaussianGenerator.getGaussian(mean, _levelSettings.levelSizeVariance, rand);
float sizeY = (float)GaussianGenerator.getGaussian(mean, _levelSettings.levelSizeVariance, rand);
int finalSizeX = (int)Math.Round(Mathf.Clamp(sizeX, _levelSettings.minLevelSize, _levelSettings.maxLevelSize));
int finalSizeY = (int)Math.Round(Mathf.Clamp(sizeY, _levelSettings.minLevelSize, _levelSettings.maxLevelSize));
return(new IntVector2(finalSizeX, finalSizeY));
}
public void Mutate(ref GaussianGenerator rnd, float learningRate)
{
foreach (Layer layer in model.layers)
{
for (int iWeight = 0; iWeight < layer.weights.Length; iWeight++)
{
layer.weights[iWeight] += rnd.NextFloat1() * learningRate;
}
}
PrepareCache();
}
public void GaussianTest()
{
Random _rndu = new Random((uint)UnityEngine.Random.Range(0, int.MaxValue));
GaussianGenerator _rndn = new GaussianGenerator(_rndu);
for (int iCnt = 0; iCnt < 100; iCnt++)
{
float val = _rndn.NextFloat1();
Debug.Log(val);
Debug.Assert(!float.IsNaN(val));
}
}
public static double Beta()
{
GaussianGenerator generator = new GaussianGenerator(0, 1);
double x = generator.Next();
x *= x;
x *= 0.5;
double y = generator.Next();
y *= y;
y *= 0.5;
return(x / (x + y));
}
/// <summary>
/// Random Gamma-distribution number generation
/// based on Marsaglia's Simple Method (2000).
/// </summary>
///
public static double Random(double d, double c)
{
var g = new GaussianGenerator(0, 1, Generator.Random.Next());
// References:
//
// - Marsaglia, G. A Simple Method for Generating Gamma Variables, 2000
//
while (true)
{
// 2. Generate v = (1+cx)^3 with x normal
double x, t, v;
do
{
x = g.Next();
t = (1.0 + c * x);
v = t * t * t;
} while (v <= 0);
// 3. Generate uniform U
double U = Accord.Math.Random.Generator.Random.NextDouble();
// 4. If U < 1-0.0331*x^4 return d*v.
double x2 = x * x;
if (U < 1 - 0.0331 * x2 * x2)
{
return(d * v);
}
// 5. If log(U) < 0.5*x^2 + d*(1-v+log(v)) return d*v.
if (Math.Log(U) < 0.5 * x2 + d * (1.0 - v + Math.Log(v)))
{
return(d * v);
}
// 6. Goto step 2
}
}
/// <summary>
/// Static generator initialization.
/// </summary>
static Util()
{
Gaussian = new GaussianGenerator(0, 1);
Uniform = new UniformGenerator(new Range(0, 1));
}
public void Start()
{
_rndg = new GaussianGenerator(new Random((uint)UnityEngine.Random.Range(0, int.MaxValue)));
}
