Disable FAST_COMPILE on windows again

[aseyboldt]

Revert the workaround to use FAST_COMPILE on Windows. This has unacceptable performance implications, and setting no-c++11-narrowing should be enough.
A better solution would be to fix the code that is generated by theano.

[codecov]

Codecov Report

Merging #4011 into master will increase coverage by 0.01%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #4011      +/-   ##
==========================================
+ Coverage   86.65%   86.66%   +0.01%     
==========================================
  Files          88       88              
  Lines       14090    14087       -3     
==========================================
- Hits        12210    12209       -1     
+ Misses       1880     1878       -2     

Impacted Files	Coverage Δ
pymc3/__init__.py	100.00% <100.00%> (+4.76%)	⬆️

[twiecki]

Why is this in draft status? I think we might want to get this in for 3.7.3?

[michaelosthege]

I ran tests on my machine. It seems to be mostly fine. Unfortunately pytest didn't tell me which tests failed:

=========================================================================================================== test session starts ============================================================================================================
platform win32 -- Python 3.6.10, pytest-5.4.3, py-1.8.1, pluggy-0.13.1
rootdir: C:\Users\osthege\Repos\pymc3-dev, inifile: setup.cfg
collected 3593 items

test_coords.py .                                                                                                                                                                                                                      [  0%]
test_data_container.py .............                                                                                                                                                                                                  [  0%]
test_dist_math.py ......s......F                                                                                                                                                                                                      [  0%]
test_distribution_defaults.py ..........                                                                                                                                                                                              [  1%]
test_distributions.py ..........X............................F..........................................................................................................                                                              [  5%]
test_distributions_random.py ................s.....x................................................................................................................................................................................. [ 10%]
..................................................................................................................................................................................................................................... [ 17%]
..................................................................................................................................................................................................................................... [ 23%]
..................................................................................................................................F.....................s...xF
(pm3-dev2) C:\Users\osthege\Repos\pymc3-dev\pymc3\tests>

They could also be unrelated as I didn't run the tests from the master branch before..

The one from test_dist_math.py is TypeError: data type "float128" not understood.

From test_distributions.py - also unrelated:

test_distributions.py ..........X............................F..........................................................................................................                                                              [100%]

================================================================================================================= FAILURES =================================================================================================================
______________________________________________________________________________________________________ TestMatchesScipy.test_weibull _______________________________________________________________________________________________________

self = <pymc3.tests.test_distributions.TestMatchesScipy object at 0x000002118046A710>

    @pytest.mark.xfail(
        condition=(theano.config.floatX == "float32"),
        reason="Fails on float32 due to inf issues",
    )
    def test_weibull(self):
        self.pymc3_matches_scipy(
            Weibull,
            Rplus,
            {"alpha": Rplusbig, "beta": Rplusbig},
>           lambda value, alpha, beta: sp.exponweib.logpdf(value, 1, alpha, scale=beta),
        )

test_distributions.py:968:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
test_distributions.py:538: in pymc3_matches_scipy
    self.check_logp(model, value, domain, paramdomains, logp, decimal=decimal)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

self = <pymc3.tests.test_distributions.TestMatchesScipy object at 0x000002118046A710>, model = <pymc3.model.Model object at 0x00000211FFA2EBE0>, value = value, domain = <pymc3.tests.test_distributions.Domain object at 0x00000211F9FD8C18>
paramdomains = {'alpha': <pymc3.tests.test_distributions.Domain object at 0x00000211F9FE07F0>, 'beta': <pymc3.tests.test_distributions.Domain object at 0x00000211F9FE07F0>}
logp_reference = <function TestMatchesScipy.pymc3_matches_scipy.<locals>.logp at 0x00000211FEE7E950>, decimal = 6

    def check_logp(
        self, model, value, domain, paramdomains, logp_reference, decimal=None
    ):
        domains = paramdomains.copy()
        domains["value"] = domain
        logp = model.fastlogp
        for pt in product(domains, n_samples=100):
            pt = Point(pt, model=model)
            if decimal is None:
                decimal = select_by_precision(float64=6, float32=3)
            assert_almost_equal(
>               logp(pt), logp_reference(pt), decimal=decimal, err_msg=str(pt)
            )
E           AssertionError:
E           Arrays are not almost equal to 6 decimals
E           {'alpha': array(20.), 'beta': array(0.9), 'value': array(100.)}
E           Mismatched elements: 1 / 1 (100%)
E           Max absolute difference: 3.86856262e+25
E           Max relative difference: 4.70326902e-16
E            x: array(-8.225263e+40)
E            y: array(-8.225263e+40)

test_distributions.py:551: AssertionError

test_distributions_random.py:

================================================================================================================= FAILURES =================================================================================================================
______________________________________________________________________________________________ TestScalarParameterSamples.test_beta_binomial _______________________________________________________________________________________________

self = <pymc3.tests.test_distributions_random.TestScalarParameterSamples object at 0x0000017AF7DDB518>

    def test_beta_binomial(self):
        pymc3_random_discrete(pm.BetaBinomial, {'n': Nat, 'alpha': Rplus, 'beta': Rplus},
>                             ref_rand=self._beta_bin)

test_distributions_random.py:639:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
test_distributions_random.py:73: in pymc3_random_discrete
    e = ref_rand(size=size, **pt)
test_distributions_random.py:642: in _beta_bin
    return st.binom.rvs(n, st.beta.rvs(a=alpha, b=beta, size=size))
..\..\..\..\appdata\local\continuum\miniconda3\envs\pm3-dev2\lib\site-packages\scipy\stats\_distn_infrastructure.py:2969: in rvs
    return super(rv_discrete, self).rvs(*args, **kwargs)
..\..\..\..\appdata\local\continuum\miniconda3\envs\pm3-dev2\lib\site-packages\scipy\stats\_distn_infrastructure.py:980: in rvs
    vals = self._rvs(*args)
..\..\..\..\appdata\local\continuum\miniconda3\envs\pm3-dev2\lib\site-packages\scipy\stats\_discrete_distns.py:43: in _rvs
    return self._random_state.binomial(n, p, self._size)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

>   ???
E   TypeError: Cannot cast array data from dtype('int64') to dtype('int32') according to the rule 'safe'

mtrand.pyx:3347: TypeError

[twiecki]

@michaelosthege I would strongly expect those to be unrelated, which OS did you test this under? Travis seems fine.

[michaelosthege]

@twiecki Yes, I think they are unrelated. Windows.
Makes a great argument in favor of #3954 though.