pdfs: introduce UniPdf (uniform distribution), includes tests

pybayes/__init__.py

@@ -4,5 +4,5 @@
 
 """TODO: documentation"""
 
-from pdfs import Pdf, GaussPdf
+from pdfs import Pdf, UniPdf, GaussPdf
 from kalman import Kalman

pybayes/numpywrap.py

@@ -8,4 +8,4 @@
 
 from numpy import any as np_any, array, asarray, diag, dot, dot as dotvv, ndarray
 from numpy.linalg import cholesky, inv, slogdet
-from numpy.random import normal
+from numpy.random import normal, uniform

pybayes/numpywrap.pyx

@@ -11,7 +11,7 @@
 from numpy cimport import_array, int, npy_intp, NPY_DOUBLE, PyArray_EMPTY, PyArray_ISCARRAY_RO, PyArray_ISFARRAY_RO
 from numpy import any as np_any, array, asarray, diag, empty
 from numpy.linalg import cholesky, slogdet
-from numpy.random import normal
+from numpy.random import normal, uniform
 
 cimport tokyo as t
 

pybayes/pdfs.pxd

@@ -21,6 +21,11 @@ cdef class Pdf:
     cpdef ndarray sample(self)
 
 
+cdef class UniPdf(Pdf):
+
+    cdef public double a, b
+
+
 cdef class GaussPdf(Pdf):
 
     cdef public ndarray mu

pybayes/pdfs.py

@@ -4,6 +4,8 @@
 
 """Probability density functions"""
 
+from math import log
+
 from numpywrap import *
 
 
@@ -31,10 +33,47 @@ def sample(self):
         raise NotImplementedError("Derived classes must implement this function")
 
 
+class UniPdf(Pdf):
+    """Simple uniform single-dimensional probability density function
+
+    .. math: f(x|a, b) = \THETA {x-a} \THETA {b-x} {1} \over {b-a}  TODO
+    """
+
+    def __init__(self, a, b):
+        """Initialise uniform distribution with left point a and right point b
+
+        a must be greater that b
+        """
+        if b <= a:
+            raise ValueError("b must be grater than a")
+        self.a = float(a)
+        self.b = float(b)
+
+    def shape(self):
+        return (1,)
+
+    def mean(self):
+        return array([(self.a+self.b)/2.])
+
+    def variance(self):
+        return array([((self.b-self.a)**2)/12.])
+
+    def eval_log(self, x):
+        if x is None:  # cython-specific, but wont hurt in python
+            raise ValueError("x must be numpy.ndarray")
+        x0 = x[0]
+        if x0 <= self.a or x0 >= self.b:
+            return float('-inf')
+        return -log(self.b-self.a)
+
+    def sample(self):
+        return uniform(self.a, self.b, self.shape())
+
+
 class GaussPdf(Pdf):
     """Unconditional Gaussian (normal) probability density function
 
-    .. math: f(x|\mu,b) \propto \exp(-(x-\mu)'R^{-1}(x-\mu))
+    .. math: f(x|\mu, R) \propto \exp(-(x-\mu)'R^{-1}(x-\mu))
     """
 
     def __init__(self, mean=array([0]), covariance=array([[1]])):

pybayes/tests/test_pdfs.py

@@ -4,7 +4,7 @@
 
 """Tests for pdfs"""
 
-from math import exp
+from math import exp, log
 import unittest as ut
 
 import numpy as np
@@ -32,6 +32,36 @@ def test_abstract_methods(self):
         self.assertRaises(NotImplementedError, self.pdf.sample)
 
 
+class TestUniPdf(ut.TestCase):
+    """Test uniform pdf"""
+
+    def setUp(self):
+        self.uni = pb.UniPdf(-10., 20.)
+
+    def test_init(self):
+        self.assertEqual(type(self.uni), pb.UniPdf)
+
+    def test_invalid_init(self):
+        self.assertRaises(ValueError, pb.UniPdf, 1.0, 0.5)
+
+    def test_shape(self):
+        self.assertEqual(self.uni.shape(), (1,))
+
+    def test_mean(self):
+        self.assertTrue(np.all(self.uni.mean() == np.array([5.])))
+
+    def test_variance(self):
+        self.assertTrue(np.all(self.uni.variance() == np.array(75.)))
+
+    def test_eval_log(self):
+        self.assertEqual(self.uni.eval_log(np.array([-10.1])), float('-inf'))
+        self.assertEqual(self.uni.eval_log(np.array([12.547])), log(1./30.))
+        self.assertEqual(self.uni.eval_log(np.array([-10.1])), float('-inf'))
+
+    def test_sample(self):
+        pass
+
+
 class TestGaussPdf(ut.TestCase):
     """Test Gaussian pdf"""