np.unique no longer works with fixed float and string arrays in Python 3

[astrofrog]

In Python 2, the following code used to work, but no longer in Python 3:

In [6]: x = np.array([1.1, 'a'], dtype=np.object)

In [7]: np.unique(x)
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-7-9a478241c930> in <module>()
----> 1 np.unique(x)

/Users/tom/Library/Python/3.4/lib/python/site-packages/numpy/lib/arraysetops.py in unique(ar, return_index, return_inverse, return_counts)
    193         aux = ar[perm]
    194     else:
--> 195         ar.sort()
    196         aux = ar
    197     flag = np.concatenate(([True], aux[1:] != aux[:-1]))

TypeError: unorderable types: float() > str()

Is this intended behavior, and if so, are there any suggested alternatives in this situation?

[juliantaylor]

its not specific to numpy, its normal python3 behaviour for objects:

In [1]: sorted([1.1, u'a'])
TypeError: unorderable types: str() < float()

[juliantaylor]

hm we have a pr somewhere that uses hashes for unique of object arrays somewhere ..

[juliantaylor]

hm no just an issue, #641
closing as duplicate