There is a new bug in master brunch, scikit-learn/sklearn/utils/fixes.py
in line 406: if np_version < (1, 12, 0): , an error occured, TypeError: unorderable types: str() < int().
It's because in the function :

def _parse_version(version_string):
    version = []
    for x in version_string.split('.'):
        try:
            version.append(int(x))
        except ValueError:
            # x may be of the form dev-1ea1592
            version.append(x)
    return tuple(version)

However， my numpy version is：1.12.0b1, this funciotn return a tuple(1,12,'0b1') , but in line 406, It is compared with integer type, so the error occured.
I just in a simple way to solve it. I change the code to if np_version < (1, 12, '0'): in line 406, just change the third tuple index to string type.
Of course, this is not a common method.