add sync_rayleight_perlin

priorConditionedAnnealing/noise.py

@@ -100,7 +100,6 @@ class SDE_Noise():
         noise = th.bmm(th.bmm(latent_sde, self.exploration_matrices), chol)
         return noise.squeeze(dim=1)
 
-
 class Perlin_Noise():
     def __init__(self, known_shape=None, scale=0.1, octave=1):
         self.known_shape = known_shape
@@ -126,7 +125,7 @@ class Perlin_Noise():
         self.index = 0
         self.noise = PerlinNoise(octaves=self.octave)
 
-class P_Perlin_Noise():
+class Async_Perlin_Noise():
     def __init__(self, known_shape=None, scale=0.1, octave=1):
         self.known_shape = known_shape
         self.scale = scale
@@ -220,4 +219,30 @@ class Rayleigh_Perlin_Noise():
     def reset(self):
         self.index = 0
         self.scales = np.random.rayleigh(scale=self.sigma, size=np.prod(self.known_shape[:-1]))
-        self.noise = PerlinNoise(octaves=1)
+        self.noise = PerlinNoise(octaves=1)
+
+class Sync_Rayleigh_Perlin_Noise():
+    def __init__(self, known_shape=None, sigma=0.1):
+        self.known_shape = known_shape
+        self.sigma = sigma
+        self.magic = PI  # Axis offset, should be (kinda) irrational
+        # We want to genrate samples, that approx ~N(0,1)
+        self.normal_factor = PI/20
+        self.clear_cache_every = 128
+        self.reset()
+
+    def __call__(self, shape=None):
+        if shape == None:
+            shape = self.known_shape
+        self.index += 1
+        noise = [self.noise([self.index*self.scale, self.magic+(2*a)]) / self.normal_factor
+                 for a in range(shape[-1])]
+        if self.index % self.clear_cache_every == 0:
+            self.noise.cache = {}
+        return th.Tensor(noise)
+
+    def reset(self):
+        self.index = 0
+        self.scale = np.random.rayleigh(scale=self.sigma, size=(1,))[0]
+        self.noise = PerlinNoise(octaves=1)
+