Added stats.logpdf. Prepare for release.

creating_a_release.rst

@@ -1,7 +1,10 @@
 Steps to Create Release
 =======================
+
 * update filterpy/filterpy/__init__.py with the version number.
 
+* update filterpy/filterpy/changelog.txt with the changes for this release.
+
 * 'rm *' in dist
 
 * If necessary, edit filterpy/docs/index.rst to add any classes. Add .rst file for those new classes to the /docs subdirectories.

docs/stats/stats.rst

@@ -21,6 +21,10 @@ to Bayesian filters.
 
 -----
 
+.. autofunction:: logpdf
+
+-----
+
 .. autofunction:: multivariate_gaussian
 
 -----

filterpy/__init__.py

@@ -14,4 +14,4 @@
 for more information.
 """
 
-__version__ = "0.1.4"
+__version__ = "0.1.5"

filterpy/changelog.txt

@@ -4,6 +4,10 @@ Version 0.1.5
 * Fix #53: UKF rts_smoother does not use residual functions
 * Fixed #54: Comments in multivariate_multiply incorrectly called the
   covariance the mean. 
+* Added logpdf to stats. Computes logpdf - mostly a wrapper around
+  stats.multivariate_normal.logpdf as older versions of that function 
+  do not support the allow_singular keyword. But it also flattens out the
+  vectors for you so you do not have to do anything special with column vectors.
 
 Version 0.1.4
 =============

filterpy/kalman/UKF.py

@@ -19,11 +19,11 @@
 
 from filterpy.common import dot3
 from filterpy.kalman import unscented_transform
+from filterpy.stats import logpdf
 import math
 import numpy as np
 from numpy import eye, zeros, dot, isscalar, outer
 from scipy.linalg import inv, cholesky
-from scipy.stats import multivariate_normal
 
 class UnscentedKalmanFilter(object):
     # pylint: disable=too-many-instance-attributes
@@ -359,8 +359,16 @@ def update(self, z, R=None, UT=None, hx_args=()):
         self.x = self.x + dot(self.K, self.y)
         self.P = self.P - dot3(self.K, Pz, self.K.T)
 
-        self.log_likelihood = multivariate_normal.logpdf(
-            x=self.y, mean=np.zeros(len(self.y)), cov=Pz, allow_singular=True)
+        self.log_likelihood = logpdf(self.y, np.zeros(len(self.y)), Pz)
+
+
+    def cross_variance(self, x, z, sigmas_f, sigmas_h):
+        Pxz = zeros((sigmas_f.shape[1], sigmas_h.shape[1]))
+        N = sigmas_f.shape[0]
+        for i in range(N):
+            dx = self.residual_x(sigmas_f[i], x)
+            dz =  self.residual_z(sigmas_h[i], z)
+            Pxz += self.Wc[i] * outer(dx, dz)
 
 
     @property

filterpy/kalman/kalman_filter.py

@@ -17,11 +17,11 @@
 
 from __future__ import (absolute_import, division, unicode_literals)
 from filterpy.common import setter, setter_1d, setter_scalar, dot3
+from filterpy.stats import logpdf
 import math
 import numpy as np
 from numpy import dot, zeros, eye, isscalar, shape
 import scipy.linalg as linalg
-from scipy.stats import multivariate_normal
 
 
 class KalmanFilter(object):
@@ -198,12 +198,7 @@ def update(self, z, R=None, H=None):
         self._S = S
         self._K = K
 
-        # compute log likelihood
-        mean = np.asarray(dot(H, x)).flatten()
-        flatz = np.asarray(z).flatten()
-        self.log_likelihood = multivariate_normal.logpdf(
-             flatz, mean, cov=S, allow_singular=True)
-
+        self.log_likelihood = logpdf(z, dot(H, x), S)
 
 
     def update_correlated(self, z, R=None, H=None):
@@ -270,11 +265,7 @@ def update_correlated(self, z, R=None, H=None):
         self._K = K
 
         # compute log likelihood
-        mean = np.asarray(dot(H, x)).flatten()
-        flatz = np.asarray(z).flatten()
-
-        self.log_likelihood = multivariate_normal.logpdf(
-             flatz, mean, cov=S, allow_singular=True)
+        self.log_likelihood = logpdf(z, dot(H, x), S)
 
 
     def test_matrix_dimensions(self, z=None, H=None, R=None, F=None, Q=None):
@@ -897,12 +888,10 @@ def update(x, P, z, R, H=None, return_all=False):
         I_KH = np.array(1 - KH)
     P = dot3(I_KH, P, I_KH.T) + dot3(K, R, K.T)
 
-    # compute log likelihood
-    mean = np.array(dot(H, x)).flatten()
-    flatz = np.asarray(z).flatten()
-    log_likelihood = multivariate_normal.logpdf(flatz, mean, cov=S, allow_singular=True)
 
     if return_all:
+        # compute log likelihood
+        log_likelihood = logpdf(z, dot(H, x), S)
         return x, P, y, K, S, log_likelihood
     else:
         return x, P

filterpy/kalman/tests/test_enkf.py

@@ -134,8 +134,8 @@ def fx(x, dt):
 
 if __name__ == '__main__':
     DO_PLOT = True
-    test_circle ()
-    #test_1d_const_vel()
+    #test_circle ()
+    test_1d_const_vel()
 
 
 #test_noisy_1d()