BUG/TST: io: fix redundant if-condition in arffread.py

[distroinfinity]

Reference issue

#11864

What does this implement/fix?

Tried to improve the elif "yy" condition

Additional information

[tylerjereddy]

Adding some debug prints I see:

__________________________________________________________________________________________________________ TestDateAttribute.test_yeari_attribute __________________________________________________________________________________________________________
TypeError: invalid type promotion

The above exception was the direct cause of the following exception:
scipy/io/arff/tests/test_arffread.py:219: in test_yeari_attribute
    assert_array_equal(self.data["attr_yeari"], expected)
E   DeprecationWarning: elementwise comparison failed; this will raise an error in the future.
------------------------------------------------------------------------------------------------------------------- Captured stdout call -------------------------------------------------------------------------------------------------------------------
actual: ['1999-01-31' '2004-12-01' '2017-04-28' '2000-09-10' '2013-11-30'
 '2031-10-15']
expected: ['0099' '0004' '0017' '0000' '0013' '0031']
datetime64[D] datetime64[Y]

[tylerjereddy]

Are "years" the smallest units you're expecting to get back here? If I switch to days it is a little closer:

actual: ['1999-01-31' '2004-12-01' '2017-04-28' '2000-09-10' '2013-11-30'
 '2031-10-15']
expected: ['0099-01-31' '0004-12-01' '0017-04-28' '0000-09-10' '0013-11-30'
 '0031-10-15']
datetime64[D] datetime64[D]

Looks like 31 is getting converted to 2031, but I suppose we're expecting only the two-digit year to be extracted unambiguously here.

[tylerjereddy]

This seems to be what happens under the hood:

>>> datetime.datetime.strptime('31-10-15', '%y-%m-%d')
datetime.datetime(2031, 10, 15, 0, 0)

[tylerjereddy]

I could be wrong, but the prospects for returning a datetime object with the century somehow stripped out are maybe not great cc @eric-wieser ?

The question of what we expect to get back here in SciPy is actually not immediately clear to me, but perhaps it would actually match the century restoration behavior of the std lib?