Removed use of scipy's mahalanobis function

It creates deprecation warning if you dont' pass in 1D arrays.
Filterpy's mahalanobis function is more forgiving, so switched
to using it instead.

Author: rlabbe

filterpy/kalman/tests/test_fm.py

@@ -21,8 +21,8 @@
 import numpy as np
 import matplotlib.pyplot as plt
 from filterpy.kalman import FadingKalmanFilter
+from filterpy.stats import mahalanobis
 from pytest import approx
-from scipy.spatial.distance import mahalanobis as scipy_mahalanobis
 
 DO_PLOT = False
 def test_noisy_1d():
@@ -59,7 +59,7 @@ def test_noisy_1d():
 
         # test mahalanobis
         a = np.zeros(f.y.shape)
-        maha = scipy_mahalanobis(a, f.y, f.SI)
+        maha = mahalanobis(a, f.y, f.S)
         assert f.mahalanobis == approx(maha)
         print(z, maha, f.y, f.S)
         assert maha < 4
@@ -87,4 +87,4 @@ def test_noisy_1d():
 
 if __name__ == "__main__":
     DO_PLOT = True
-    test_noisy_1d()
+    test_noisy_1d()

filterpy/kalman/tests/test_kf.py

@@ -24,8 +24,8 @@
 from pytest import approx
 from filterpy.kalman import KalmanFilter, update, predict, batch_filter
 from filterpy.common import Q_discrete_white_noise, kinematic_kf, Saver
+from filterpy.stats import mahalanobis
 from scipy.linalg import block_diag, norm
-from scipy.spatial.distance import mahalanobis as scipy_mahalanobis
 
 DO_PLOT = False
 
@@ -121,7 +121,7 @@ def test_noisy_1d():
 
         # test mahalanobis
         a = np.zeros(f.y.shape)
-        maha = scipy_mahalanobis(a, f.y, f.SI)
+        maha = mahalanobis(a, f.y, f.S)
         assert f.mahalanobis == approx(maha)
 
 
@@ -233,7 +233,7 @@ def test_noisy_11d():
 
         # test mahalanobis
         a = np.zeros(f.y.shape)
-        maha = scipy_mahalanobis(a, f.y, f.SI)
+        maha = mahalanobis(a, f.y, f.S)
         assert f.mahalanobis == approx(maha)
 
     # now do a batch run with the stored z values so we can test that
@@ -443,7 +443,7 @@ def test_steadystate():
         cv.update_steadystate([i])
         # test mahalanobis
         a = np.zeros(cv.y.shape)
-        maha = scipy_mahalanobis(a, cv.y, cv.SI)
+        maha = mahalanobis(a, cv.y, cv.S)
         assert cv.mahalanobis == approx(maha)
 
 

filterpy/stats/tests/test_stats.py

@@ -23,13 +23,30 @@
 import numpy as np
 from numpy.linalg import inv
 import scipy
-from scipy.spatial.distance import mahalanobis as scipy_mahalanobis
-from filterpy.stats import norm_cdf, multivariate_gaussian, logpdf, mahalanobis
+from scipy.spatial.distance import mahalanobis as _scipy_mahalanobis
+from filterpy.stats import (norm_cdf, multivariate_gaussian, logpdf,
+                            mahalanobis)
 from scipy import linalg
 
 
 ITERS = 10000
 
+
+def scipy_mahalanobis(x, mean, cov):
+    # scipy 1.9 will not accept scalars as input, so force the correct
+    # behavior so we don't get deprecation warnings or exceptions
+
+    def validate_vector(u):
+        u = np.asarray(u).squeeze()
+        # Ensure values such as u=1 and u=[1] still return 1-D arrays.
+        u = np.atleast_1d(u)
+        return u
+
+    x = validate_vector(x)
+    mean = validate_vector(mean)
+    return _scipy_mahalanobis(x, mean, cov)
+
+
 def test_mahalanobis():
     global a, b, S
     # int test
@@ -92,7 +109,6 @@ def test_multivariate_gaussian():
     with warnings.catch_warnings():
         warnings.simplefilter("ignore")
 
-
         # test that we treat lists and arrays the same
         mean= (0, 0)
         cov=[[1, .5], [.5, 1]]
@@ -294,6 +310,8 @@ def covariance_3d_plot_test():
     plot_3d_covariance(mu, C, alpha=.4, std=3, limit_xyz=True, ax=ax)
 
 if __name__ == "__main__":
+    test_multivariate_gaussian()
+    test_mahalanobis()
     test_logpdf2()
     covariance_3d_plot_test()
     plt.figure()