[Numpy-discussion] bug in lexsort with two different dtypes?

[Charles R Harris]

On 6/26/07, Charles R Harris <charlesr.harris at gmail.com> wrote:
>
>
>
> On 6/26/07, Tom Denniston <tom.denniston at alum.dartmouth.org> wrote:
> >
> > In [1]: intArr1 = numpy.array([ 0, 1, 2,-2,-1, 5,-5,-5])
> > In [2]: intArr2 = numpy.array([1,1,1,2,2,2,3,4])
> > In [3]: charArr = numpy.array(['a','a','a','b','b','b','c','d'])
> >
> > Here I sort two int arrays.  As expected intArr2 dominates intArr1 but
> > the items with the same intArr2 values are sorted forwards according
> > to intArr1
> > In [6]: numpy.lexsort((intArr1, intArr2))
> > Out[6]: array([0, 1, 2, 3, 4, 5, 6, 7])
> >
> > This,  however, looks like a bug to me.  Here I sort an int array and
> > a str array.  As expected charArray dominates intArr1 but the items
> > with the same charArray values are sorted *backwards* according to
> > intArr1
> > In [5]: numpy.lexsort((intArr1, charArr))
> > Out[5]: array([2, 1, 0, 5, 4, 3, 6, 7])
> >
> > Is this a bug or am I missing something?
>
>
> Looks like a bug.
>
> In [12]: numpy.argsort([charArr], kind='m')
> Out[12]: array([[2, 1, 0, 5, 4, 3, 6, 7]])
>
> In [13]: numpy.argsort([intArr2], kind='m')
> Out[13]: array([[0, 1, 2, 3, 4, 5, 6, 7]])
>
> Both of these are stable sorts, and since the elements are in order should
> return [[0, 1, 2, 3, 4, 5, 6, 7]]. Actually, I think they should return [0,
> 1, 2, 3, 4, 5, 6, 7], I'm not sure why the returned array is 2D and I
> suspect that is a bug also. As to why the string array sorts incorrectly, I
> am not sure. It could be that the sort isn't stable, there could be a stride
> error, or the comparison is returning wrong values. My bet is on the first
> being the case.
>

Nevermind the 2D thingee, that was pilot error in changing lexsort to
argsort, charArr should not be in a list:

In [25]: numpy.argsort(charArr, kind='m', axis=0)
Out[25]: array([2, 1, 0, 5, 4, 3, 6, 7])

Works just fine.

Chuck
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/numpy-discussion/attachments/20070626/3fcb350a/attachment.html>