BUG: Test failures from BLAS-flavour testing for 1.13 in conda-forge #20472

h-vetinari · 2024-04-14T11:51:57Z

Collecting the various test failures (new and existing) from conda-forge/scipy-feedstock#273, I made a distinction whether something is affecting the main BLAS flavours (e.g. the conda-forge default, or other frequently used implementations like MKL), or regarding the severity of the failure (e.g. crash vs. minor tolerance violation) when opening new issues.

Here's a list of already existing or new stand-alone issues:

[various] BUG: test failure in test_x0_equals_Mb with bicgstab #15533
[linux-aarch64 + OpenBLAS] TST: Test failures with openblas >=0.3.24 in conda-forge on aarch #19210
[win-64 + MKL] BUG: TestNNLS.test_nnls_inner_loop_case1 fails on windows with MKL #20470
[win-64 + MKL / linux-aarch64 + netlib] BUG: TestEig.test_falker fails on windows + MKL as well as linux + aarch64 + netlib #20471
[win-64 + OpenBLAS] TST: new test failures (mostly crashes) with scipy 1.12 and OpenBLAS 0.3.26 #20271

A note on PyPy

On linux-64, PyPy 7.3.15 hangs completely, while on osx & win, there are merely additional test failures (compared to CPython).

Other failures

linux-64 + blis / osx + accelerate: minor tolerance violation in test_x0_working[tfqmr]

____________________________ test_x0_working[tfqmr] ____________________________
[gw1] linux -- Python 3.9.19 $PREFIX/bin/python
$PREFIX/lib/python3.9/site-packages/scipy/sparse/linalg/_isolve/tests/test_iterative.py:507: in test_x0_working
    assert norm(A @ x - b) <= 2e-6*norm(b)
E   assert 3.5355292454346426e-06 <= (2e-06 * 1.240298651028214)
E    +  where 3.5355292454346426e-06 = norm(((array([[3.68514493, 2.22433683, 2.80424149, 2.06524296, 2.82772901,\n        1.70134349, 2.98806253, 2.88379072, 1.9578..., 2.29515724, 2.77591422, 1.55001596, 1.93969785,\n        1.88091968, 2.24894423, 2.49791763, 1.66944017, 2.12310297]]) @ array([ -1.22142941, -23.52880604,  -4.72398545, -21.36850239,\n       -12.68971238,  -0.96740459,  32.8587725 , -11.40286367,\n        10.993623  ,  30.93083318])) - array([0.55127821, 0.22828434, 0.09135873, 0.07249963, 0.11693987,\n       0.74246847, 0.72216603, 0.1607452 , 0.15471589, 0.18022921])))
E    +  and   1.240298651028214 = norm(array([0.55127821, 0.22828434, 0.09135873, 0.07249963, 0.11693987,\n       0.74246847, 0.72216603, 0.1607452 , 0.15471589, 0.18022921]))
        A          = array([[3.68514493, 2.22433683, 2.80424149, 2.06524296, 2.82772901,
        1.70134349, 2.98806253, 2.88379072, 1.9578..., 2.29515724, 2.77591422, 1.55001596, 1.93969785,
        1.88091968, 2.24894423, 2.49791763, 1.66944017, 2.12310297]])
        b          = array([0.55127821, 0.22828434, 0.09135873, 0.07249963, 0.11693987,
       0.74246847, 0.72216603, 0.1607452 , 0.15471589, 0.18022921])
        info       = 0
        kw         = {'atol': 0, 'rtol': 1e-06}
        n          = 10
        rng        = Generator(PCG64) at 0x7FBA278194A0
        solver     = <function tfqmr at 0x7fba34af4b80>
        x          = array([ -1.22142941, -23.52880604,  -4.72398545, -21.36850239,
       -12.68971238,  -0.96740459,  32.8587725 , -11.40286367,
        10.993623  ,  30.93083318])
        x0         = array([0.63685093, 0.46836296, 0.0677924 , 0.52832101, 0.49467215,
       0.68333247, 0.3392967 , 0.57867322, 0.25729156, 0.60952728])

linux-64 + MKL: single-element tolerance violation in TestKMean.test_krandinit[numpy]

_______________________ TestKMean.test_krandinit[numpy] ________________________
[gw0] linux -- Python 3.11.8 $PREFIX/bin/python
$PREFIX/lib/python3.11/site-packages/scipy/cluster/tests/test_vq.py:353: in test_krandinit
    xp_assert_close(orig_cov, init_cov, atol=1e-2)
        data       = array([[-2.2 ,  1.17, -1.63,  1.69, -2.04,  4.38, -3.09,  0.95, -1.7 ,
         4.79, -1.68,  0.68, -2.26,  3.34, -2.2...75, -2.25,  1.71, -3.28,
         3.38, -1.74,  0.88, -2.41,  1.92, -2.24,  1.19, -2.48,  1.06,
        -1.68, -0.62]])
        datas      = [array([[-2.2 ,  1.17],
       [-1.63,  1.69],
       [-2.04,  4.38],
       [-3.09,  0.95],
       [-1.7 ,  4.79],
  ...5, -2.25,  1.71, -3.28,
         3.38, -1.74,  0.88, -2.41,  1.92, -2.24,  1.19, -2.48,  1.06,
        -1.68, -0.62]])]
        init       = array([[-1.55614868, -0.5955572 , -1.14277154, ...,  1.06323592,
        -1.4042778 ,  4.2915102 ],
       [-1.5199948...012,  2.05841291],
       [-1.82097104,  0.48772385, -4.30463642, ..., -0.02417925,
        -2.36226902,  1.46849236]])
        init_cov   = array([[ 2.52765840e-01, -2.12853305e-01, -1.70382413e-01,
        -3.35383740e-01,  3.08041712e-01, -5.99226107e-01,
...-8.46350354e-01,
         9.73843820e-01,  7.60819946e-02, -2.40444555e-01,
        -1.05667721e+00,  1.94581156e+00]])
        k          = 1000000
        orig_cov   = array([[ 2.53401111e-01, -2.13683333e-01, -1.70653333e-01,
        -3.35983333e-01,  3.09443333e-01, -6.01860000e-01,
...-8.46272222e-01,
         9.76020000e-01,  7.60555556e-02, -2.42171111e-01,
        -1.06294000e+00,  1.95102333e+00]])
        rng        = Generator(PCG64) at 0x7F0AD379AB20
        self       = <scipy.cluster.tests.test_vq.TestKMean object at 0x7f0ae2c23c50>
        xp         = <module 'numpy' from '/home/conda/feedstock_root/build_artifacts/scipy-split_1713003735196/_test_env_placehold_placeho...ehold_placehold_placehold_placehold_placehold_placehold_placehold_plac/lib/python3.11/site-packages/numpy/__init__.py'>
$PREFIX/lib/python3.11/contextlib.py:81: in inner
    return func(*args, **kwds)
E   AssertionError: 
E   Not equal to tolerance rtol=1e-07, atol=0.01
E   
E   Mismatched elements: 1 / 400 (0.25%)
E   Max absolute difference: 0.01098377
E   Max relative difference: 0.16674294
E    x: array([[ 2.534011e-01, -2.136833e-01, -1.706533e-01, -3.359833e-01,
E            3.094433e-01, -6.018600e-01,  1.971678e-01, -1.761122e-01,
E           -3.160811e-01,  3.538044e-01, -3.322333e-02, -4.241900e-01,...
E    y: array([[ 2.527658e-01, -2.128533e-01, -1.703824e-01, -3.353837e-01,
E            3.080417e-01, -5.992261e-01,  1.964842e-01, -1.761007e-01,
E           -3.145892e-01,  3.528359e-01, -3.218698e-02, -4.244195e-01,...
        args       = (<function assert_allclose.<locals>.compare at 0x7f0ad3823560>, array([[ 2.53401111e-01, -2.13683333e-01, -1.70653333e...8.46350354e-01,
         9.73843820e-01,  7.60819946e-02, -2.40444555e-01,
        -1.05667721e+00,  1.94581156e+00]]))
        func       = <function assert_array_compare at 0x7f0af4933100>
        kwds       = {'equal_nan': True, 'err_msg': '', 'header': 'Not equal to tolerance rtol=1e-07, atol=0.01', 'verbose': True}
        self       = <contextlib._GeneratorContextManager object at 0x7f0af495ca50>

osx-64 + MKL + PyPy: in test_cossin[True-40-12-20-float32]

Unclear why this only shows up in this combination (osx, MKL, PyPy); previously (for 1.12) also test_cossin[False-40-12-20-float32] failed.

______________________ test_cossin[True-40-12-20-float32] ______________________
[gw2] darwin -- Python 3.9.18 $PREFIX/bin/python
$PREFIX/lib/pypy3.9/site-packages/scipy/linalg/tests/test_decomp_cossin.py:61: in test_cossin
    assert_allclose(u, u2, rtol=0., atol=10*np.finfo(dtype_).eps)
        _          = array([], shape=(0, 0), dtype=float32)
        cs         = array([[0.9776542 , 0.        , 0.        , ..., 0.        , 0.        ,
        0.        ],
       [0.        , 0.93...       ],
       [0.        , 0.        , 0.        , ..., 0.        , 0.        ,
        0.        ]], dtype=float32)
        cs2        = array([[0.9776543 , 0.        , 0.        , ..., 0.        , 0.        ,
        0.        ],
       [0.        , 0.93...       ],
       [0.        , 0.        , 0.        , ..., 0.        , 0.        ,
        0.        ]], dtype=float32)
        dtype_     = <class 'numpy.float32'>
        m          = 40
        p          = 12
        q          = 20
        rng        = Generator(PCG64) at 0x7F80C2F5AEE0
        swap_sign  = True
        u          = array([[-0.06412254, -0.169602  ,  0.3865385 , ...,  0.        ,
         0.        ,  0.        ],
       [-0.217334 ...6],
       [ 0.        ,  0.        ,  0.        , ...,  0.15757903,
        -0.14554036, -0.19666514]], dtype=float32)
        u2         = array([[-0.06412242,  0.1696023 ,  0.3865385 , ...,  0.        ,
         0.        ,  0.        ],
       [-0.2173338...6],
       [ 0.        ,  0.        ,  0.        , ...,  0.15757942,
        -0.14554064, -0.1966648 ]], dtype=float32)
        vh         = array([[ 0.13631816, -0.20160665, -0.25449932, ...,  0.        ,
         0.        ,  0.        ],
       [ 0.0383219... ],
       [ 0.        ,  0.        ,  0.        , ...,  0.13668793,
         0.01040455, -0.26890454]], dtype=float32)
        vh2        = array([[ 0.13631816, -0.20160665, -0.25449932, ...,  0.        ,
         0.        ,  0.        ],
       [ 0.0383219... ],
       [ 0.        ,  0.        ,  0.        , ...,  0.13668793,
         0.01040455, -0.26890454]], dtype=float32)
        x          = array([[-0.02367539,  0.40119368, -0.05368597, ...,  0.17337501,
         0.37582624,  0.01585775],
       [ 0.0132678...1],
       [ 0.15149911,  0.1761488 , -0.08100508, ..., -0.14594388,
        -0.35172257, -0.04196982]], dtype=float32)
$PREFIX/lib/pypy3.9/contextlib.py:79: in inner
    return func(*args, **kwds)
E   AssertionError: 
E   Not equal to tolerance rtol=0, atol=1.19209e-06
E   
E   Mismatched elements: 161 / 1600 (10.1%)
E   Max absolute difference: 1.2539275
E   Max relative difference: 2.0001893
E    x: array([[-0.064123, -0.169602,  0.386539, ...,  0.      ,  0.      ,
E            0.      ],
E          [-0.217334, -0.533708, -0.280909, ...,  0.      ,  0.      ,...
E    y: array([[-0.064122,  0.169602,  0.386539, ...,  0.      ,  0.      ,
E            0.      ],
E          [-0.217334,  0.533708, -0.280909, ...,  0.      ,  0.      ,...
        args       = (<function assert_allclose.<locals>.compare at 0x00007f80b7eb2660>, array([[-0.06412254, -0.169602  ,  0.3865385 , ......],
       [ 0.        ,  0.        ,  0.        , ...,  0.15757942,
        -0.14554064, -0.1966648 ]], dtype=float32))
        func       = <function assert_array_compare at 0x00007f80c7d3a340>
        kwds       = {'equal_nan': True, 'err_msg': '', 'header': 'Not equal to tolerance rtol=0, atol=1.19209e-06', 'verbose': True}
        self       = <contextlib._GeneratorContextManager object at 0x00007f80c7d483a0>

PyPy (everywhere): test_dunder_ror[dok_matrix]

_________________________ test_dunder_ror[dok_matrix] __________________________
[gw0] darwin -- Python 3.9.18 $PREFIX/bin/python
$PREFIX/lib/pypy3.9/site-packages/scipy/sparse/tests/test_dok.py:163: in test_dunder_ror
    assert d.__ror__(d) == Asp.__ror__(d)
E   AttributeError: 'dict' object has no attribute '__ror__'. Did you mean: '__ior__'?
        Asp        = <3x3 sparse matrix of type '<class 'numpy.float64'>'
	with 2 stored elements in Dictionary Of Keys format>
        d          = {(0, 1): 1, (0, 2): 2}

Fixed ✅

In #20474

{linux-64,win-64} + blis: minor (absolute) tolerance violation in TestRegularGridInterpolator.test_derivatives

_________________ TestRegularGridInterpolator.test_derivatives _________________
[gw1] linux -- Python 3.9.19 $PREFIX/bin/python
$PREFIX/lib/python3.9/site-packages/scipy/interpolate/tests/test_rgi.py:139: in test_derivatives
    assert_allclose(interp(sample, nu=(0, 1, 1, 0)),
        interp     = <scipy.interpolate._rgi.RegularGridInterpolator object at 0x7fba3082fa60>
        points     = [(0.0, 0.5, 1.0), (0.0, 0.5, 1.0), (0.0, 0.5, 1.0), (0.0, 0.5, 1.0)]
        sample     = array([[0.1 , 0.1 , 1.  , 0.9 ],
       [0.2 , 0.1 , 0.45, 0.8 ],
       [0.5 , 0.5 , 0.5 , 0.5 ]])
        self       = <scipy.interpolate.tests.test_rgi.TestRegularGridInterpolator object at 0x7fba3082f610>
        values     = array([[[[0.0000e+00, 5.0000e+02, 1.0000e+03],
         [5.0000e+01, 5.5000e+02, 1.0500e+03],
         [1.0000e+02, 6....1100e+02, 1.0110e+03],
         [6.1000e+01, 5.6100e+02, 1.0610e+03],
         [1.1100e+02, 6.1100e+02, 1.1110e+03]]]])
$PREFIX/lib/python3.9/contextlib.py:79: in inner
    return func(*args, **kwds)
E   AssertionError: 
E   Not equal to tolerance rtol=1e-07, atol=1e-12
E   
E   Mismatched elements: 1 / 3 (33.3%)
E   Max absolute difference: 1.02318154e-12
E   Max relative difference: inf
E    x: array([-1.023182e-12,  0.000000e+00, -9.094947e-13])
E    y: array([0, 0, 0])
        args       = (<function assert_allclose.<locals>.compare at 0x7fba23d8e8b0>, array([-1.02318154e-12,  0.00000000e+00, -9.09494702e-13]), array([0, 0, 0]))
        func       = <function assert_array_compare at 0x7fba3c55cb80>
        kwds       = {'equal_nan': True, 'err_msg': '', 'header': 'Not equal to tolerance rtol=1e-07, atol=1e-12', 'verbose': True}
        self       = <contextlib._GeneratorContextManager object at 0x7fba3c5650a0>

osx-64 + accelerate: tolerance violation in TestInterpN.test_matrix_input[quintic]

____________________ TestInterpN.test_matrix_input[quintic] ____________________
[gw1] darwin -- Python 3.10.14 $PREFIX/bin/python
$PREFIX/lib/python3.10/site-packages/scipy/interpolate/tests/test_rgi.py:986: in test_matrix_input
    assert_allclose(v1, v2)
        method     = 'quintic'
        sample     = array([[[0.86763282, 0.44517443],
        [0.02918942, 0.10719824],
        [0.19625915, 0.23823634],
        [0.82451...3, 0.34864109],
        [0.48048149, 0.64392643],
        [0.62162643, 0.76119301],
        [0.28665106, 0.49238521]]])
        self       = <scipy.interpolate.tests.test_rgi.TestInterpN object at 0x119037370>
        v1         = array([[0.4789054 , 0.3514018 , 0.54012216, 0.54447361, 0.47604678,
        0.51217751, 0.04425903],
       [0.8209919...778, 0.68049855],
       [0.95113819, 0.548359  , 0.69364055, 0.32331236, 0.5971776 ,
        0.07371108, 0.82078431]])
        v2         = array([[0.47890549, 0.35140192, 0.54012223, 0.5444737 , 0.47604662,
        0.51217757, 0.04425903],
       [0.8209918...821, 0.68049851],
       [0.95113806, 0.54835903, 0.69364065, 0.32331228, 0.59717755,
        0.07371097, 0.82078419]])
        values     = matrix([[0.03931146, 0.43177652, 0.85427585, 0.42232848, 0.08844911,
         0.05669501, 0.10327285],
        [0.4853...2, 0.45127308],
        [0.65087142, 0.80930517, 0.28296286, 0.27338179, 0.38573719,
         0.42160812, 0.93216543]])
        x          = array([0. , 0.4, 0.8, 1.2, 1.6, 2. ])
        y          = array([0.        , 0.16666667, 0.33333333, 0.5       , 0.66666667,
       0.83333333, 1.        ])
$PREFIX/lib/python3.10/contextlib.py:79: in inner
    return func(*args, **kwds)
E   AssertionError: 
E   Not equal to tolerance rtol=1e-07, atol=0
E   
E   Mismatched elements: 15 / 21 (71.4%)
E   Max absolute difference: 4.27814137e-07
E   Max relative difference: 1.98036236e-06
E    x: array([[0.478905, 0.351402, 0.540122, 0.544474, 0.476047, 0.512178,
E           0.044259],
E          [0.820992, 0.284413, 1.280366, 0.476941, 0.861279, 0.216028,...
E    y: array([[0.478905, 0.351402, 0.540122, 0.544474, 0.476047, 0.512178,
E           0.044259],
E          [0.820992, 0.284413, 1.280366, 0.476941, 0.861279, 0.216028,...
        args       = (<function assert_allclose.<locals>.compare at 0x1278ff010>, array([[0.4789054 , 0.3514018 , 0.54012216, 0.54447361, 0...21, 0.68049851],
       [0.95113806, 0.54835903, 0.69364065, 0.32331228, 0.59717755,
        0.07371097, 0.82078419]]))
        func       = <function assert_array_compare at 0x1136a2ef0>
        kwds       = {'equal_nan': True, 'err_msg': '', 'header': 'Not equal to tolerance rtol=1e-07, atol=0', 'verbose': True}
        self       = <contextlib._GeneratorContextManager object at 0x1136b8f40>

osx-64 + accelerate: tolerance violation in TestRBFInterpolatorNeighbors20.test_pickleable (flaky)

________________ TestRBFInterpolatorNeighbors20.test_pickleable ________________
[gw1] darwin -- Python 3.9.18 $PREFIX/bin/python
$PREFIX/lib/pypy3.9/site-packages/scipy/interpolate/tests/test_rbfinterp.py:418: in test_pickleable
    assert_array_equal(yitp1, yitp2)
        interp     = <scipy.interpolate._rbfinterp.RBFInterpolator object at 0x00007fd326723910>
        self       = <scipy.interpolate.tests.test_rbfinterp.TestRBFInterpolatorNeighbors20 object at 0x00007fd33e91ee90>
        seq        = <scipy.stats._qmc.Halton object at 0x00007fd326722fa8>
        x          = array([[0.      ],
       [1.5     ],
       [0.75    ],
       [2.25    ],
       [0.375   ],
       [1.875   ],
    ...75],
       [0.609375],
       [2.109375],
       [1.359375],
       [2.859375],
       [0.140625],
       [1.640625]])
        xitp       = array([[0.890625 ],
       [2.390625 ],
       [0.515625 ],
       [2.015625 ],
       [1.265625 ],
       [2.765625 ]...      [1.4296875],
       [2.9296875],
       [0.0703125],
       [1.5703125],
       [0.8203125],
       [2.3203125]])
        y          = array([ 0.        ,  0.24413581, -0.44285432,  0.27466724, -0.9722152 ,
        0.29864165,  0.02432428,  0.2240337 , ...131898, -0.12065703,  0.2441372 , -0.66695964,
        0.28883559,  0.18663159,  0.19056618, -0.77998605,  0.27856231])
        yitp1      = array([-0.23846054,  0.25702357, -0.81124363,  0.29531296,  0.13262317,
        0.20380022, -0.9961497 ,  0.29755   , ...415359, -0.55418691,  0.28230372,  0.21859066,
        0.18102502, -0.48531143,  0.26388737, -0.33687091,  0.26616824])
        yitp2      = array([-0.23846054,  0.25702357, -0.81124363,  0.29531296,  0.13262317,
        0.20380022, -0.9961497 ,  0.29755   , ...415359, -0.55418691,  0.28230372,  0.21859066,
        0.18102502, -0.48531143,  0.26388737, -0.33687091,  0.26616824])
$PREFIX/lib/pypy3.9/contextlib.py:79: in inner
    return func(*args, **kwds)
E   AssertionError: 
E   Arrays are not equal
E   
E   Mismatched elements: 3 / 50 (6%)
E   Max absolute difference: 5.55111512e-17
E   Max relative difference: 2.04146e-16
E    x: array([-0.238461,  0.257024, -0.811244,  0.295313,  0.132623,  0.2038  ,
E          -0.99615 ,  0.29755 , -0.019889,  0.230791, -0.516801,  0.279867,
E           0.227767,  0.178071, -0.172756,  0.251327, -0.406818,  0.271919,...
E    y: array([-0.238461,  0.257024, -0.811244,  0.295313,  0.132623,  0.2038  ,
E          -0.99615 ,  0.29755 , -0.019889,  0.230791, -0.516801,  0.279867,
E           0.227767,  0.178071, -0.172756,  0.251327, -0.406818,  0.271919,...
        args       = (<built-in function eq>, array([-0.23846054,  0.25702357, -0.81124363,  0.29531296,  0.13262317,
        0.20380022, -...15359, -0.55418691,  0.28230372,  0.21859066,
        0.18102502, -0.48531143,  0.26388737, -0.33687091,  0.26616824]))
        func       = <function assert_array_compare at 0x00007fd336574660>
        kwds       = {'err_msg': '', 'header': 'Arrays are not equal', 'strict': False, 'verbose': True}
        self       = <contextlib._GeneratorContextManager object at 0x00007fd33655fa98>

win-64 + blis: minor tolerance violation in Test_SVDS_PROPACK.test_svds_parameter_k_which[LM-5]

_____________ Test_SVDS_PROPACK.test_svds_parameter_k_which[LM-5] _____________
[gw1] win32 -- Python 3.10.14 %PREFIX%\python.exe
..\_test_env\lib\site-packages\scipy\sparse\linalg\_eigen\tests\test_svds.py:269: in test_svds_parameter_k_which
    _check_svds(A, k, *res, which=which, atol=8e-10)
        A          = array([[0.63696169, 0.26978671, 0.04097352, 0.01652764, 0.81327024,
        0.91275558, 0.60663578, 0.72949656, 0.5436..., 0.96792619, 0.0147063 , 0.86364009, 0.98119504,
        0.95721018, 0.14876401, 0.97262881, 0.88993556, 0.82237383]])
        k          = 5
        res        = (array([[-0.32771492,  0.4158033 , -0.53137496,  0.02426458,  0.31783075],
       [-0.39239685, -0.02251468, -0.063087...54 ,  0.21588656,  0.29722489,  0.31843038,
         0.31819307,  0.28881893,  0.46509902,  0.29898348,  0.31066731]]))
        rng        = Generator(PCG64) at 0x27FE5935000
        self       = <scipy.sparse.linalg._eigen.tests.test_svds.Test_SVDS_PROPACK object at 0x0000027FD06F29B0>
        which      = 'LM'
..\_test_env\lib\site-packages\scipy\sparse\linalg\_eigen\tests\test_svds.py:52: in _check_svds
    assert_allclose(uh_u, np.identity(k), atol=atol, rtol=rtol)
        A          = array([[0.63696169, 0.26978671, 0.04097352, 0.01652764, 0.81327024,
        0.91275558, 0.60663578, 0.72949656, 0.5436..., 0.96792619, 0.0147063 , 0.86364009, 0.98119504,
        0.95721018, 0.14876401, 0.97262881, 0.88993556, 0.82237383]])
        A_rebuilt  = array([[ 0.62664866,  0.26143156,  0.11583711,  0.10946253,  0.90237439,
         0.86426285,  0.54138144,  0.62698622...6753, -0.05990525,  0.81456418,  0.85739979,
         1.00404725,  0.21265842,  1.03219384,  0.92104378,  0.79572735]])
        atol       = 8e-10
        check_svd  = True
        check_usvh_A = False
        k          = 5
        m          = 10
        n          = 10
        rtol       = 1e-07
        s          = array([0.90017941, 0.97420967, 1.26999904, 1.59849819, 5.71090105])
        u          = array([[-0.32771492,  0.4158033 , -0.53137496,  0.02426458,  0.31783075],
       [-0.39239685, -0.02251468, -0.0630873...84 , -0.12697327, -0.09277105,  0.34919585],
       [-0.26457914, -0.13503641, -0.15837359,  0.53003248,  0.42841613]])
        uh_u       = array([[ 1.00000000e+00, -3.27833058e-14, -1.38947663e-15,
         2.42856942e-13, -1.48403292e-10],
       [-3.27833...3.47129694e-11],
       [-1.48403292e-10,  9.92425943e-10,  5.40502283e-10,
         3.47129694e-11,  1.00000000e+00]])
        vh         = array([[-0.42927645, -0.02385317,  0.21368543,  0.47797589, -0.42344063,
         0.30012128,  0.03406212,  0.19590089...554 ,  0.21588656,  0.29722489,  0.31843038,
         0.31819307,  0.28881893,  0.46509902,  0.29898348,  0.31066731]])
        which      = 'LM'
..\_test_env\lib\contextlib.py:79: in inner
    return func(*args, **kwds)
E   AssertionError: 
E   Not equal to tolerance rtol=1e-07, atol=8e-10
E   
E   Mismatched elements: 2 / 25 (8%)
E   Max absolute difference: 9.92425943e-10
E   Max relative difference: 1.80300219e-13
E    x: array([[ 1.000000e+00, -3.278331e-14, -1.389477e-15,  2.428569e-13,
E           -1.484033e-10],
E          [-3.278331e-14,  1.000000e+00,  1.498714e-14, -6.832333e-13,...
E    y: array([[1., 0., 0., 0., 0.],
E          [0., 1., 0., 0., 0.],
E          [0., 0., 1., 0., 0.],...
        args       = (<function assert_allclose.<locals>.compare at 0x0000027FE74E1F30>, array([[ 1.00000000e+00, -3.27833058e-14, -1.38947...,
       [0., 1., 0., 0., 0.],
       [0., 0., 1., 0., 0.],
       [0., 0., 0., 1., 0.],
       [0., 0., 0., 0., 1.]]))
        func       = <function assert_array_compare at 0x0000027FC28DAB00>
        kwds       = {'equal_nan': True, 'err_msg': '', 'header': 'Not equal to tolerance rtol=1e-07, atol=8e-10', 'verbose': True}
        self       = <contextlib._GeneratorContextManager object at 0x0000027FC28CE470>

The text was updated successfully, but these errors were encountered:

A slight tolerance violation was reported on conda-forge in scipy#20472

A small tolerance violation reported on conda-forge in scipy#20472

An exact equality was reported as flaky on conda-forge in scipy#20472 The tolerance violations are of the order of fp noise (< 2e-16), and I don't think that pickling/unpickling guarantees bit-to-bit compatibility. In principle, this may invoke recalculations and those may be subject to fp noise. So I think it's OK to only require allclose(atol=eps) instead of exact equality.

ev-br · 2024-04-14T13:38:32Z

#20474 has what I think are straightforward tolerance bumps in RGI and RBF.

The cossin failure needs a more careful look. Off the cuff, it looks like some signs differ. So maybe it's a matter of the factorization only being unique up to consistent sign changes in the multipliers.

An exact equality was reported as flaky on conda-forge in scipy#20472 The tolerance violations are of the order of fp noise (< 2e-16), and I don't think that pickling/unpickling guarantees bit-to-bit compatibility. In principle, this may invoke recalculations and those may be subject to fp noise. So I think it's OK to only require allclose(atol=eps) instead of exact equality.

h-vetinari · 2024-04-15T01:12:30Z

The failure in test_x0_working[tfqmr] also showed up on osx-64 + accelerate now...

osx-64 + accelerate: minor tolerance violation in test_x0_working[tfqmr]

____________________________ test_x0_working[tfqmr] ____________________________
[gw0] darwin -- Python 3.11.8 $PREFIX/bin/python
$PREFIX/lib/python3.11/site-packages/scipy/sparse/linalg/_isolve/tests/test_iterative.py:507: in test_x0_working
    assert norm(A @ x - b) <= 2e-6*norm(b)
E   assert 2.5193553673686355e-06 <= (2e-06 * 1.240298651028214)
E    +  where 2.5193553673686355e-06 = norm(((array([[3.68514493, 2.22433683, 2.80424149, 2.06524296, 2.82772901,\n        1.70134349, 2.98806253, 2.88379072, 1.9578..., 2.29515724, 2.77591422, 1.55001596, 1.93969785,\n        1.88091968, 2.24894423, 2.49791763, 1.66944017, 2.12310297]]) @ array([ -1.22141632, -23.52877158,  -4.72398748, -21.36850005,\n       -12.68970431,  -0.96739246,  32.85873976, -11.40287273,\n        10.9936288 ,  30.93080707])) - array([0.55127821, 0.22828434, 0.09135873, 0.07249963, 0.11693987,\n       0.74246847, 0.72216603, 0.1607452 , 0.15471589, 0.18022921])))
E    +  and   1.240298651028214 = norm(array([0.55127821, 0.22828434, 0.09135873, 0.07249963, 0.11693987,\n       0.74246847, 0.72216603, 0.1607452 , 0.15471589, 0.18022921]))
        A          = array([[3.68514493, 2.22433683, 2.80424149, 2.06524296, 2.82772901,
        1.70134349, 2.98806253, 2.88379072, 1.9578..., 2.29515724, 2.77591422, 1.55001596, 1.93969785,
        1.88091968, 2.24894423, 2.49791763, 1.66944017, 2.12310297]])
        b          = array([0.55127821, 0.22828434, 0.09135873, 0.07249963, 0.11693987,
       0.74246847, 0.72216603, 0.1607452 , 0.15471589, 0.18022921])
        info       = 0
        kw         = {'atol': 0, 'rtol': 1e-06}
        n          = 10
        rng        = Generator(PCG64) at 0x1220C8120
        solver     = <function tfqmr at 0x10c87ad40>
        x          = array([ -1.22141632, -23.52877158,  -4.72398748, -21.36850005,
       -12.68970431,  -0.96739246,  32.85873976, -11.40287273,
        10.9936288 ,  30.93080707])
        x0         = array([0.63685093, 0.46836296, 0.0677924 , 0.52832101, 0.49467215,
       0.68333247, 0.3392967 , 0.57867322, 0.25729156, 0.60952728])

A slight tolerance violation was reported on conda-forge in scipy#20472

A small tolerance violation reported on conda-forge in scipy#20472

An exact equality was reported as flaky on conda-forge in scipy#20472 The tolerance violations are of the order of fp noise (< 2e-16), and I don't think that pickling/unpickling guarantees bit-to-bit compatibility. In principle, this may invoke recalculations and those may be subject to fp noise. So I think it's OK to only require allclose(atol=eps) instead of exact equality.

h-vetinari added the defect A clear bug or issue that prevents SciPy from being installed or used as expected label Apr 14, 2024

h-vetinari mentioned this issue Apr 14, 2024

TEST: 1.13.x + blas variants conda-forge/scipy-feedstock#273

Draft

ev-br added a commit to ev-br/scipy that referenced this issue Apr 14, 2024

TST: interpolate: bump test tolerance in TestRGI::test_derivatives

e2d3153

A slight tolerance violation was reported on conda-forge in scipy#20472

ev-br added a commit to ev-br/scipy that referenced this issue Apr 14, 2024

TST: interpolate: bump test tolerance in TestInterpN

ca71394

A small tolerance violation reported on conda-forge in scipy#20472

ev-br mentioned this issue Apr 14, 2024

TST: tolerance bumps for the conda-forge builds #20474

Merged

thalassemia mentioned this issue Apr 17, 2024

BLD: Accelerate wheels for macOS 14+ #20510

Merged

h-vetinari mentioned this issue May 2, 2024

Rebuild for numpy 2.0 conda-forge/scipy-feedstock#274

Merged

tylerjereddy pushed a commit to tylerjereddy/scipy that referenced this issue May 2, 2024

TST: interpolate: bump test tolerance in TestRGI::test_derivatives

3c3ac73

A slight tolerance violation was reported on conda-forge in scipy#20472

tylerjereddy pushed a commit to tylerjereddy/scipy that referenced this issue May 2, 2024

TST: interpolate: bump test tolerance in TestInterpN

5006ef1

A small tolerance violation reported on conda-forge in scipy#20472

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

BUG: Test failures from BLAS-flavour testing for 1.13 in conda-forge #20472

BUG: Test failures from BLAS-flavour testing for 1.13 in conda-forge #20472

h-vetinari commented Apr 14, 2024 •

edited

ev-br commented Apr 14, 2024

h-vetinari commented Apr 15, 2024

BUG: Test failures from BLAS-flavour testing for 1.13 in conda-forge #20472

BUG: Test failures from BLAS-flavour testing for 1.13 in conda-forge #20472

Comments

h-vetinari commented Apr 14, 2024 • edited

A note on PyPy

Other failures

Fixed ✅

In #20474

ev-br commented Apr 14, 2024

h-vetinari commented Apr 15, 2024

h-vetinari commented Apr 14, 2024 •

edited