sklearn issue#2560

In [10]: a = np.arange(20).reshape(10,2).astype(np.float)

In [12]: a[3,1] = np.nan

In [14]: a[5,0] = np.nan

In [15]: a
Out[15]: 
array([[  0.,   1.],
       [  2.,   3.],
       [  4.,   5.],
       [  6.,  nan],
       [  8.,   9.],
       [ nan,  11.],
       [ 12.,  13.],
       [ 14.,  15.],
       [ 16.,  17.],
       [ 18.,  19.]])

In [16]: from sklearn.utils import array2d

In [17]: X = array2d(a, force_all_finite=False)

In [18]: X
Out[18]: 
array([[  0.,   1.],
       [  2.,   3.],
       [  4.,   5.],
       [  6.,  nan],
       [  8.,   9.],
       [ nan,  11.],
       [ 12.,  13.],
       [ 14.,  15.],
       [ 16.,  17.],
       [ 18.,  19.]])

In [20]: def _get_mask(X, value_to_mask):
   ....:         """Compute the boolean mask X == missing_values."""
   ....:         if value_to_mask == "NaN" or np.isnan(value_to_mask):
   ....:                 return np.isnan(X)
   ....:         else:
   ....:                 return X == value_to_mask
   ....:     

In [21]: mask = _get_mask(X, "NaN")

In [24]: masked_X = np.ma.masked_array(X, mask=mask)

In [27]: median_masked = np.ma.median(masked_X, axis=0)

In [28]: median_masked
Out[28]: 
masked_array(data = [  8.  11.],
             mask = False,
       fill_value = 1e+20)

In [29]: mean_masked = np.ma.mean(masked_X, axis=0)

In [30]: mean_masked
Out[30]: 
masked_array(data = [8.88888888888889 10.333333333333334],
             mask = [False False],
       fill_value = 1e+20)