API: bool dtype casts to object arrays with np.array

[jseabold]

Bug? Did this used to work?

np.array(pandas.DataFrame.from_dict(dict(a=np.random.rand(50), b=np.random.randint(2, size=50).astype(bool))))

We would expect that this casts to a plain float ndarray.

See statsmodels/statsmodels#880

[josef-pkt]

>>> np.find_common_type([bool, float], [])
dtype('float64')

[jreback]

this returns an object dtype and has since at least 0.10.0 (could be before as well).

[jreback]

This is not hard to change, but has a couple of internal implications....so will need some investigation

Currently

In [27]: df = pandas.DataFrame.from_dict(dict(a=np.random.rand(10), b=np.random.randint(2, size=10).astype(bool)))

In [28]: df
Out[28]: 
          a      b
0  0.906756   True
1  0.780954   True
2  0.057409  False
3  0.486914   True
4  0.130978  False
5  0.257315  False
6  0.123006  False
7  0.783898  False
8  0.904467   True
9  0.915809  False

In [29]: df.dtypes
Out[29]: 
a    float64
b       bool
dtype: object

In [30]: df.values
Out[30]: 
array([[0.9067555725633248, True],
       [0.7809537562875581, True],
       [0.05740922820129213, False],
       [0.4869138045816368, True],
       [0.13097846147217107, False],
       [0.2573150347933001, False],
       [0.12300624903001778, False],
       [0.7838978280636173, False],
       [0.9044667038591694, True],
       [0.9158090069212313, False]], dtype=object)

You can easily do this, however

In [4]: df.astype(float).values
Out[4]: 
array([[ 0.35721166,  1.        ],
       [ 0.87632264,  1.        ],
       [ 0.88853538,  1.        ],
       [ 0.36573411,  1.        ],
       [ 0.21589354,  1.        ],

[jreback]

this is as expected.