domain check for a < b in stats.truncnorm

[akahard2dj]

I fix and a PR of the bug issued by @chrisb83.
I changed the condition formula for py_test from a!=b and a<b to
np.bitwise_and(np.not_equal(a, b), np.less(a, b)) in _argcheck.

Closes #9169

[chrisb83]

Less (<) implies not_equal

[akahard2dj]

I used the numpy function as a one-to-one correspondence during the argument checking.
As you already fully knows, the != symbol is np.not_equal, the < symbol is np.less.
If so, does it mean that you can use only one np.less function in this section?

[ilayn]

return a < b does the same thing.

[akahard2dj]

OK, I will reflect that mention.
But, I used the np.less function because it fails in py_test when using a<b condition.

bshape     = [5, 4, 3, 2]
dist       = 'truncnorm'
distfunc   = <scipy.stats._continuous_distns.truncnorm_gen object at 0x6ffef15b710>
k          = 1
loc        = array([0., 0.])
nargs      = 2
scale      = array([[1.],
       [1.],
       [1.]])
shape_args = (0.1, 2.0)
shape_only = False
shp        = (5, 1, 1, 1)

scipy/stats/tests/test_continuous_basic.py:258:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
scipy/stats/tests/common_tests.py:300: in check_rvs_broadcast
    sample = distfunc.rvs(*allargs)
scipy/stats/_distn_infrastructure.py:936: in rvs
    cond = logical_and(self._argcheck(*args), (scale >= 0))
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

self = <scipy.stats._continuous_distns.truncnorm_gen object at 0x6ffef15b710>
a = array([[[[0.1, 0.1],
         [0.1, 0.1],
         [0.1, 0.1]],

        [[0.1, 0.1],
         [0.1, 0.1],
         [0....1, 0.1],
         [0.1, 0.1],
         [0.1, 0.1]],

        [[0.1, 0.1],
         [0.1, 0.1],
         [0.1, 0.1]]]])
b = array([[[[2., 2.],
         [2., 2.],
         [2., 2.]],

        [[2., 2.],
         [2., 2.],
         [2., 2.]],

...
        [[2., 2.],
         [2., 2.],
         [2., 2.]],

        [[2., 2.],
         [2., 2.],
         [2., 2.]]]])

 def _argcheck(self, a, b):
        self.a = a
        self.b = b
        self._nb = _norm_cdf(b)
        self._na = _norm_cdf(a)
        self._sb = _norm_sf(b)
        self._sa = _norm_sf(a)
        self._delta = np.where(self.a > 0,
                               -(self._sb - self._sa),
                               self._nb - self._na)
        self._logdelta = np.log(self._delta)
>       return a != b and a < b
E       ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

[ilayn]

It should be only return a < b. Equality is ruled out by strict inequality < anyways.

If you want to combine bool statements then it should read as (a != b) & (a < b). But here not needed.

[akahard2dj]

OK, I understood an expression of equality and comparison operator.
I will use the comparison operator < as your advised.
My worries were about the py_test as well as the comparison operator.
I used the numpy function(np.less, np.not_equal, np.bitwise_and) to solve the section
where py_test failed in multiple condition.
As a result, I will commit the conditional statements used only by the comparison operator.

Thanks, your valuable advice @ilayn, @chrisb83.

[ilayn]

I've added Closes clause to your PR description to link to the issue for automatic closing.

[ilayn]

All green, thanks @akahard2dj @chrisb83