scipy/stats/tests/test_continuous_basic.py::test_cont_basic[500-200-ncf-arg74] test fails with IntegrationWarning

[mgorny]

scipy/stats/tests/test_continuous_basic.py::test_cont_basic[500-200-ncf-arg74] test fails in 1.7.1.

Error message:

_________________________________________________ test_cont_basic[500-200-ncf-arg74] __________________________________________________
scipy/stats/tests/test_continuous_basic.py:189: in test_cont_basic
    check_entropy(distfn, arg, distname)
        alpha      = 0.01
        arg        = (27, 27, 0.41578441799226107)
        distfn     = <scipy.stats._continuous_distns.ncf_gen object at 0x7f8939255910>
        distname   = 'ncf'
        locscale_defaults = (0, 1)
        m          = array(1.09663138)
        meths      = [<bound method rv_continuous.pdf of <scipy.stats._continuous_distns.ncf_gen object at 0x7f8939255910>>, <bound method ...f8939255910>>, <bound method rv_continuous.logsf of <scipy.stats._continuous_distns.ncf_gen object at 0x7f8939255910>>]
        n_fit_samples = 200
        rng        = RandomState(MT19937) at 0x7F88CEEBDB40
        rvs        = array([0.90888841, 0.87523216, 0.73228753, 1.33584017, 0.93548576,
       1.09504212, 1.15670442, 0.36786914, 0.791225...03, 1.73540023, 0.87035866, 0.73702581, 0.81477748,
       0.87362182, 1.58673099, 1.88455316, 0.61564337, 1.09912842])
        sm         = 1.0740685247899555
        sn         = 500
        spec_x     = {'levy_l': -0.5, 'pareto': 1.5, 'rv_histogram_instance': 5.0, 'tukeylambda': 0.3, ...}
        sv         = 0.1911827409537913
        v          = array(0.2013794)
        x          = 0.5
scipy/stats/tests/common_tests.py:93: in check_entropy
    ent = distfn.entropy(*arg)
        arg        = (27, 27, 0.41578441799226107)
        distfn     = <scipy.stats._continuous_distns.ncf_gen object at 0x7f8939255910>
        msg        = 'ncf'
scipy/stats/_distn_infrastructure.py:1254: in entropy
    place(output, cond0, self.vecentropy(*goodargs) + log(goodscale))
        args       = (array(27), array(27), array(0.41578442))
        cond0      = True
        goodargs   = [array([27]), array([27]), array([0.41578442])]
        goodscale  = array([1])
        kwds       = {}
        loc        = array(0)
        output     = array(0.)
        scale      = array(1)
        self       = <scipy.stats._continuous_distns.ncf_gen object at 0x7f8939255910>
/usr/lib/python3.8/site-packages/numpy/lib/function_base.py:2163: in __call__
    return self._vectorize_call(func=func, args=vargs)
        args       = (array([27]), array([27]), array([0.41578442]))
        excluded   = set()
        func       = <bound method rv_continuous._entropy of <scipy.stats._continuous_distns.ncf_gen object at 0x7f8939255910>>
        kwargs     = {}
        self       = <numpy.vectorize object at 0x7f8939255c10>
        vargs      = (array([27]), array([27]), array([0.41578442]))
/usr/lib/python3.8/site-packages/numpy/lib/function_base.py:2246: in _vectorize_call
    outputs = ufunc(*inputs)
        args       = (array([27]), array([27]), array([0.41578442]))
        func       = <bound method rv_continuous._entropy of <scipy.stats._continuous_distns.ncf_gen object at 0x7f8939255910>>
        inputs     = [array([27], dtype=object), array([27], dtype=object), array([0.41578441799226107], dtype=object)]
        otypes     = 'd'
        self       = <numpy.vectorize object at 0x7f8939255c10>
        ufunc      = <ufunc '_entropy (vectorized)'>
scipy/stats/_distn_infrastructure.py:2621: in _entropy
    h = integrate.quad(integ, _a, _b)[0]
        _a         = 0.0
        _b         = inf
        args       = (27, 27, 0.41578441799226107)
        integ      = <function rv_continuous._entropy.<locals>.integ at 0x7f88cba75550>
        self       = <scipy.stats._continuous_distns.ncf_gen object at 0x7f8939255910>
scipy/integrate/quadpack.py:400: in quad
    warnings.warn(msg, IntegrationWarning, stacklevel=2)
E   scipy.integrate.quadpack.IntegrationWarning: The occurrence of roundoff error is detected, which prevents 
E     the requested tolerance from being achieved.  The error may be 
E     underestimated.
        a          = 0.0
        args       = ()
        b          = inf
        epsabs     = 1.49e-08
        epsrel     = 1.49e-08
        flip       = False
        full_output = 0
        func       = <function rv_continuous._entropy.<locals>.integ at 0x7f88cba75550>
        ier        = 2
        limit      = 50
        limlst     = 50
        maxp1      = 50
        msg        = 'The occurrence of roundoff error is detected, which prevents \n  the requested tolerance from being achieved.  The error may be \n  underestimated.'
        msgs       = {80: 'A Python error occurred possibly while calling the function.', 1: 'The maximum number of subdivisions (50) has b...\n  underestimated.', 3: 'Extremely bad integrand behavior occurs at some points of the\n  integration interval.', ...}
        points     = None
        retval     = (nan, nan, 2)
        weight     = None
        wopts      = None
        wvar       = None

Scipy/Numpy/Python version information:

$  python3.8 -c 'import sys, scipy, numpy; print(scipy.__version__, numpy.__version__, sys.version_info)'
1.7.1 1.21.1 sys.version_info(major=3, minor=8, micro=11, releaselevel='final', serial=0)

[tupui]

Same... passes on master. Can you verify all three with master on your system?

[tupui]

Closing as probably related to #14517 We can reopen if you have contradicting elements later on.

[drew-parsons]

I've just seen this error in scipy 1.8.1, rebuilding on a Debian (unstable) system. It previously passed fine. The main difference is that previously the tests were run with

python3 -c "import scipy; exit( not scipy.test() )"

Now I'm trying to run them using

python3 runtests.py -n -v

(in order to get more control of which tests are run).
I can reproduce the error directly with

python3 runtests.py -n -v -- -k "test_cont_basic[500-200-ncf-arg74]"

Patch #16822 that closed #14517) only changed Schur tests. Should this Issue #14519 be reopened?

Building with gcc 12.1.0 and pythran 0.11.0.
System info:

python3.10 -c 'import sys, scipy, numpy; print(scipy.__version__, numpy.__version__, sys.version_info)'
1.8.1 1.21.5 sys.version_info(major=3, minor=10, micro=6, releaselevel='final', serial=0)

[WarrenWeckesser]

@drew-parsons, you are right, this issue is not related to #14157.

[WarrenWeckesser]

I reopened the issue, but it is still possible that the problem has been fixed (directly or indirectly) in the main development branch. For example, #15763 made significant changes to the ncf distribution. @drew-parsons, are you able to rerun the tests on your system with SciPy 1.9.0?

[drew-parsons]

Our watch script got stuck on 1.8, I hadn't seen that 1.9 is already released. I'll prepare a 1.9 build and report back.

[mdhaber]

@drew-parsons any luck here?

[drew-parsons]

I've had to put the 1.9 build on hold since it requires pymeson to build. Debian doesn't have pymeson packaged yet, but the packaging process has started, tracked at https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1017959

[mdhaber]

@drew-parsons it looks like that issue was resolved. Does the test pass now?

[drew-parsons]

debian finally got mesonpy packaged, and we now have scipy 1.10 built.
test_cont_basic[500-200-ncf-arg74] is indeed passing now.

$ python3 runtests.py -v -n -s stats -- -k "test_cont_basic[500-200-ncf-arg74]"
Running tests for scipy version:1.10.0, installed at:/usr/lib/python3/dist-packages/scipy
=============================================================================================== test session starts ===============================================================================================
platform linux -- Python 3.11.1, pytest-7.2.1, pluggy-1.0.0+repack -- /usr/bin/python3
cachedir: .pytest_cache
hypothesis profile 'default' -> database=DirectoryBasedExampleDatabase('/projects/python/build/scipy/build/test/.hypothesis/examples')
rootdir: /projects/python/build/scipy, configfile: pytest.ini
plugins: xvfb-2.0.0, remotedata-0.4.0, asyncio-0.20.3, arraydiff-0.5.0, filter-subpackage-0.1.2, flaky-3.7.0, hypothesis-6.64.0, cov-4.0.0, mpi-0.6, astropy-header-0.2.2, openfiles-0.5.0, mock-3.8.2, doctestplus-0.12.1, astropy-0.10.0
asyncio: mode=Mode.STRICT
collected 8512 items / 8511 deselected / 1 selected                                                                                                                                                               

../../tests/test_continuous_basic.py::test_cont_basic[500-200-ncf-arg74] PASSED                                                                                                                             [100%]

======================================================================================= 1 passed, 8511 deselected in 2.71s ========================================================================================