Increased the cython tolerance to avoid CI failures due to precision …

…errors
thomgrand · Jan 27, 2022 · 30720e7 · 30720e7
1 parent c84abd7
commit 30720e7
Showing 1 changed file with 6 additions and 4 deletions.
diff --git a/tests/test_cython_methods.py b/tests/test_cython_methods.py
@@ -6,6 +6,7 @@ class TestCythonMethods():
 
     @pytest.mark.parametrize("prec", [np.float32, np.float64])
     def test_2D_vec(self, prec):
+        rtol = (1e-3 if prec == np.float32 else 1e-4)
         A = np.random.normal(size=(100, 2, 2)).astype(prec)
         A[..., 1, 0] = A[..., 0, 1] #Symmetrize
 
@@ -14,7 +15,7 @@ def test_2D_vec(self, prec):
 
         expected_result = np.einsum('...x,...xy,...y->...', z1, A, z2)
         comp_result = metric_norm_matrix_2D_cython(A, z1, z2, ret_sqrt=False)
-        assert np.allclose(expected_result, comp_result, rtol=1e-4) #Single vectorized
+        assert np.allclose(expected_result, comp_result, rtol=rtol)
 
         #Broadcasted
         A = np.random.normal(size=(100, 1, 2, 2))
@@ -25,10 +26,11 @@ def test_2D_vec(self, prec):
 
         expected_result = np.einsum('...x,...xy,...y->...', z1, A, z2)
         comp_result = metric_norm_matrix_2D_cython(A, z1, z2, ret_sqrt=False)
-        assert np.allclose(expected_result, comp_result, rtol=1e-4) #Single vectorized
+        assert np.allclose(expected_result, comp_result, rtol=rtol) #Broadcasted
 
     @pytest.mark.parametrize("prec", [np.float32, np.float64])
     def test_3D_vec(self, prec):
+        rtol = (1e-3 if prec == np.float32 else 1e-4)
         A = np.random.normal(size=(100, 3, 3)).astype(prec)
         A[..., 1, 0] = A[..., 0, 1] #Symmetrize
         A[..., 2, 0] = A[..., 0, 2] 
@@ -39,7 +41,7 @@ def test_3D_vec(self, prec):
 
         expected_result = np.einsum('...x,...xy,...y->...', z1, A, z2)
         comp_result = metric_norm_matrix_3D_cython(A, z1, z2, ret_sqrt=False)
-        assert np.allclose(expected_result, comp_result, rtol=1e-4) #Single vectorized
+        assert np.allclose(expected_result, comp_result, rtol=rtol) #Single vectorized
 
         #Broadcasted
         A = np.random.normal(size=(100, 1, 3, 3))
@@ -52,6 +54,6 @@ def test_3D_vec(self, prec):
 
         expected_result = np.einsum('...x,...xy,...y->...', z1, A, z2)
         comp_result = metric_norm_matrix_3D_cython(A, z1, z2, ret_sqrt=False)
-        assert np.allclose(expected_result, comp_result, rtol=1e-4) #Single vectorized
+        assert np.allclose(expected_result, comp_result, rtol=rtol)