Voice et bot modif

venv/lib/python3.12/site-packages/numpy/linalg/_linalg.py

@@ -61,7 +61,6 @@ from numpy._core import (
     overrides,
     prod,
     reciprocal,
-    sign,
     single,
     sort,
     sqrt,
@@ -1810,7 +1809,8 @@ def svd(a, full_matrices=True, compute_uv=True, hermitian=False):
         # and related arrays to have the correct order
         if compute_uv:
             s, u = eigh(a)
-            sgn = sign(s)
+            # avoid zero sign
+            sgn = np.copysign(1.0, s)
             s = abs(s)
             sidx = argsort(s)[..., ::-1]
             sgn = np.take_along_axis(sgn, sidx, axis=-1)
@@ -2130,6 +2130,7 @@ def matrix_rank(A, tol=None, hermitian=False, *, rtol=None):
     A = asarray(A)
     if A.ndim < 2:
         return int(not all(A == 0))
+
     S = svd(A, compute_uv=False, hermitian=hermitian)
 
     if tol is None:
@@ -2137,7 +2138,7 @@ def matrix_rank(A, tol=None, hermitian=False, *, rtol=None):
             rtol = max(A.shape[-2:]) * finfo(S.dtype).eps
         else:
             rtol = asarray(rtol)[..., newaxis]
-        tol = S.max(axis=-1, keepdims=True) * rtol
+        tol = S.max(axis=-1, keepdims=True, initial=0) * rtol
     else:
         tol = asarray(tol)[..., newaxis]
 

venv/lib/python3.12/site-packages/numpy/linalg/tests/test_linalg.py

@@ -745,6 +745,12 @@ class SVDHermitianCases(HermitianTestCase, HermitianGeneralizedTestCase):
 class TestSVDHermitian(SVDHermitianCases, SVDBaseTests):
     hermitian = True
 
+    def test_singular(self):
+        x = np.array([[1, 0], [0, 0]])
+        u, _, vh = linalg.svd(x, hermitian=self.hermitian)
+        assert_allclose(u @ u.T.conj(), np.eye(2), rtol=1e-14)
+        assert_allclose(vh @ vh.T.conj(), np.eye(2), rtol=1e-14)
+
 
 class CondCases(LinalgSquareTestCase, LinalgGeneralizedSquareTestCase):
     # cond(x, p) for p in (None, 2, -2)
@@ -2440,3 +2446,22 @@ def test_vector_norm_empty():
         assert_equal(np.linalg.vector_norm(x, ord=1), 0)
         assert_equal(np.linalg.vector_norm(x, ord=2), 0)
         assert_equal(np.linalg.vector_norm(x, ord=np.inf), 0)
+
+def test_empty_matrix_rank():
+    assert_equal(matrix_rank(np.zeros((0, 0))), 0)
+    assert_equal(matrix_rank(np.zeros((0, 5))), 0)
+    assert_equal(matrix_rank(np.zeros((5, 0))), 0)
+
+    result = matrix_rank(np.zeros((0, 5, 5)))
+    assert_equal(result.shape, (0,))
+    assert_equal(result.dtype, np.intp)
+
+    result = matrix_rank(np.zeros((3, 0, 5)))
+    assert_equal(result, np.array([0, 0, 0]))
+
+    result = matrix_rank(np.zeros((2, 5, 0)))
+    assert_equal(result, np.array([0, 0]))
+
+    result = matrix_rank(np.zeros((2, 3, 0, 4)))
+    assert_equal(result.shape, (2, 3))
+    assert_equal(result, np.zeros((2, 3), dtype=np.intp))

venv/lib/python3.12/site-packages/numpy/linalg/tests/test_regression.py

@@ -7,6 +7,7 @@ import numpy as np
 from numpy import arange, array, dot, float64, linalg, transpose
 from numpy.testing import (
     assert_,
+    assert_almost_equal,
     assert_array_almost_equal,
     assert_array_equal,
     assert_array_less,
@@ -180,3 +181,10 @@ class TestRegression:
             if mismatches != 0:
                 assert False, ("unexpected result from matmul, "
                     "probably due to OpenBLAS threading issues")
+
+    def test_norm_linux_arm(self):
+        # gh-30816
+        a = np.arange(20000) / 50000
+        b = a + 1j * np.roll(np.flip(a), 12345)
+        norm = np.linalg.norm(b)
+        assert_almost_equal(norm, 46.18628948075393)