throw error when meet non ascii

[guolinke]

to fix #2226

[guolinke]

ping @StrikerRUS for review

[StrikerRUS]

Is feature_name the only model's field in which users' arbitrary strings are stored?

[guolinke]

@StrikerRUS there are some parameters with string type, e.g. data_filename, could be set by user and store in model file.

[StrikerRUS]

Maybe then add the same check for them too?

[StrikerRUS]

The check works fine with Python 3.

def test_non_ascii_names(self):
    X = np.random.random((100, 3))
    with np.testing.assert_raises_regex(lgb.basic.LightGBMError, "Do not support non-ascii*"): 
        lgb.Dataset(X, feature_name=["测试名称", "тестовая_колонка", "clé à bougie"]).construct()

But fails with Python 2.

string = '\xe6\xb5\x8b\xe8\xaf\x95\xe5\x90\x8d\xe7\xa7\xb0'

    def c_str(string):
        """Convert a Python string to C string."""
>       return ctypes.c_char_p(string.encode('utf-8'))
E       UnicodeDecodeError: 'ascii' codec can't decode byte 0xe6 in position 0: ordinal not in range(128)

So, I think it's not worth to play around with new tests for such simple check.

[jameslamb]

@StrikerRUS can this PR be merged?

[StrikerRUS]

@guolinke What do you think about enabling this check for other user's strings, like different paths as you've mentioned? #2229 (comment)

cc @jameslamb

[guolinke]

@StrikerRUS done

[StrikerRUS]

LGTM, thanks!