Added CircularGridDistribution

pyrecest/distributions/circle/circular_grid_distribution.py

@@ -0,0 +1,49 @@
+import numpy as np
+from scipy.special import sinc
+from .circular_fourier_distribution import CircularFourierDistribution
+from .abstract_circular_distribution import AbstractCircularDistribution
+from ..hypertorus.hypertoroidal_grid_distribution import HypertoroidalGridDistribution
+
+class CircularGridDistribution(HypertoroidalGridDistribution, AbstractCircularDistribution):
+    def __init__(self, grid_values, enforce_pdf_nonnegative=True):
+        if isinstance(grid_values, AbstractCircularDistribution):
+            raise ValueError("You gave a distribution as the first argument. To convert distributions to a distribution in grid representation, use .from_distribution")
+        AbstractCircularDistribution.__init__(self)
+        HypertoroidalGridDistribution.__init__(self, grid_values, enforce_pdf_nonnegative=enforce_pdf_nonnegative)
+
+    def pdf(self, xs, use_sinc=False, sinc_repetitions=5):
+        if use_sinc:
+            assert sinc_repetitions % 2 == 1
+            step_size = 2 * np.pi / len(self.grid_values)
+            _range = np.arange(-np.floor(sinc_repetitions / 2) * len(self.grid_values), np.ceil(sinc_repetitions / 2) * len(self.grid_values))
+            sinc_vals = sinc((xs / step_size).reshape(-1, 1) - _range)
+            if self.enforce_pdf_nonnegative:
+                p = np.sum(np.sqrt(self.grid_values).reshape(-1, 1) * sinc_vals, axis=1) ** 2
+            else:
+                p = np.sum(self.grid_values.reshape(-1, 1) * sinc_vals, axis=1)
+        else:
+            no_coeffs = len(self.grid_values)
+            if len(self.grid_values) % 2 == 0:
+                no_coeffs += 1
+            if self.enforce_pdf_nonnegative:
+                fd = CircularFourierDistribution.from_function_values(np.sqrt(self.grid_values), no_coeffs)
+            else:
+                fd = CircularFourierDistribution.from_function_values(self.grid_values, no_coeffs)
+            p = fd.pdf(xs)
+
+        return p
+
+    def pdf_on_grid(self, no_of_desired_gridpoints):
+        x_grid = np.linspace(0, 2 * np.pi, len(self.grid_values), endpoint=False)
+        step = len(self.grid_values) // no_of_desired_gridpoints
+        assert step == int(step), "Number of function values has to be a multiple of no_of_gridpoints"
+
+        vals = self.grid_values[::step]
+        return vals, x_grid
+
+    def trigonometric_moment(self, n):
+        no_coeffs = len(self.grid_values)
+        if len(self.grid_values) % 2 == 0:
+            no_coeffs += 1
+        fd = CircularFourierDistribution.from_function_values(self.grid_values, no_coeffs)
+        return fd.trigonometric_moment(n)