Original ticket http://projects.scipy.org/numpy/ticket/2051 on 2012-02-12 by trac user beaujolais, assigned to unknown.

Python 2.7 and 3.1 now always use the shortest decimal representation for numbers that are not exactly representable by a binary float (see [http://bugs.python.org/issue1580]).

This causes some very surprising inconsistencies with NumPy's own repr:

x = 0.1
x
0.1
a = np.array([x])
a
array([ 0.1])
a[0]
0.10000000000000001
float(a[0])
0.1
type(a[0])
<class 'numpy.float64'>

For the sake of consistency I suggest adopting the same repr in NumPy.