From 0bc1715bd81d1087668e99462c2da21e741809b9 Mon Sep 17 00:00:00 2001
From: GardevoirX <xqj.2001@outlook.com>
Date: Wed, 18 Sep 2024 16:46:19 +0200
Subject: [PATCH] Updates on the semi-positive tests

---
 src/skmatter/utils/_sparsekde.py | 17 +++++++++++------
 tests/test_neighbors.py          |  4 ++--
 2 files changed, 13 insertions(+), 8 deletions(-)

diff --git a/src/skmatter/utils/_sparsekde.py b/src/skmatter/utils/_sparsekde.py
index 3d96d2f16..ba366ca00 100644
--- a/src/skmatter/utils/_sparsekde.py
+++ b/src/skmatter/utils/_sparsekde.py
@@ -21,21 +21,26 @@ def effdim(cov):
     --------
     >>> import numpy as np
     >>> from skmatter.utils import effdim
-    >>> cov = np.array([[1, 1, 0], [1, 1, 0], [0, 0, 1]])
+    >>> cov = np.array([[25, 15, -5], [15, 18, 0], [-5, 0, 11]], dtype=np.float64)
     >>> print(round(effdim(cov), 3))
-    1.89
+    2.214
 
     References
     ----------
     https://ieeexplore.ieee.org/document/7098875
     """
     eigval = np.linalg.eigvals(cov)
+    if (lowest_eigval := np.min(eigval)) <= -np.max(cov.shape) * np.finfo(
+        cov.dtype
+    ).eps:
+        raise np.linalg.LinAlgError(
+            f"Matrix is not positive definite."
+            f"Lowest eigenvalue {lowest_eigval} is "
+            f"above numerical threshold."
+        )
+    eigval[eigval < 0.0] = 0.0
     eigval /= sum(eigval)
     eigval *= np.log(eigval)
-    if (lowest_eigval := np.min(eigval)) <= -np.max(cov.shape)*np.finfo(cov.dtype).eps:
-        raise np.linalg.LinAlgError(f"Matrix is not positive definite. Lowest eigenvalue "
-                                    f"{lowest_eigval} is above numerical threshold.")
-    eigval[eigval <= 0.] = 0.0
 
     return np.exp(-sum(eigval))
 
diff --git a/tests/test_neighbors.py b/tests/test_neighbors.py
index 14024a60a..fd6b4c0af 100644
--- a/tests/test_neighbors.py
+++ b/tests/test_neighbors.py
@@ -98,8 +98,8 @@ def test_covariance_periodic(self):
 class EffdimTests(unittest.TestCase):
     @classmethod
     def setUpClass(cls):
-        cls.cov = np.array([[1, 1, 0], [1, 1, 0], [0, 0, 1]])
-        cls.expected_effdim = 1.8898815748423097
+        cls.cov = np.array([[1, 1, 0], [1, 1.5, 0], [0, 0, 1]], dtype=np.float64)
+        cls.expected_effdim = 2.24909102090124
 
     def test_effdim(self):
         self.assertTrue(np.allclose(effdim(self.cov), self.expected_effdim))