[SciPy-user] scipy.stats.scoreatpercentile BUG ? (n, 1) shaped arrays versus (n, )

eleanor e.howick@irl.cri...
Wed Mar 5 21:10:56 CST 2008


I've just tracked problems in my code down to this unexpected behaviour

>>> import scipy
>>> d = scipy.arange(1,100)
>>> scipy.stats.scoreatpercentile(d,50)
50

That's correct
Then shuffle the data

>>> scipy.random.shuffle(d)
>>> scipy.stats.scoreatpercentile(d,50)
50

Everything's still fine

Now try it with an array where d.shape = (99,1)

>>> d2 = d.reshape(-1,1)
>>> d2.shape
(99, 1)
>>> scipy.stats.scoreatpercentile(d2,50)
array([82])

What the!!!

>>> d2[49]
array([82])

The problem seems to be that the sort in scoreatpercentile

values = np.sort(a)

dosen't sort an (n, 1) array.

I've had many problems with scipy and numpy functions not dealing well with
(n,1) arrays when they expect (n,) or is it the other way round. 

Which shape should I use with 1D arrays, to lessen my temper tantrums?

Eleanor






More information about the SciPy-user mailing list