On my machine the timings are
'5.2d' 8!:2 (%"1 <./) 100 ts&> ic ,L:0 ' D'
1.66 1.01
1.00 1.00
1.21 1.33
which makes more sense to me.
Henry Rich
> -----Original Message-----
> From: [EMAIL PROTECTED]
> [mailto:[EMAIL PROTECTED] On Behalf Of Dan Bron
> Sent: Tuesday, October 28, 2008 6:50 PM
> To: Programming forum
> Subject: Re: [Jprogramming] Index of first one in columns
>
> John wrote:
> > icol=:(] i. 1:)"1@:|:
> > Is there a better way (that avoids the transpose)?
>
> Apropos of a recent thread, you can use cut:
>
> D =: [EMAIL PROTECTED]&2 ] 400 400 NB. Test data
>
> ic0 =: (] i. 1:)"1@:|: NB. Original
> ic1 =: i.&1"1@|: NB. Variation
> ic2 =: ('';1) i.&(,1);.1 ] NB. Avoid |:
>
> ic =: ic0`ic1`ic2
>
> (-: 1&|.) ic`:0 D NB. Identical results
> 1
>
> ts =: 6!:2 , 7!:2@:]
> '5.2d' 8!:2 (%"1 <./) 100 ts&> ic ,L:0 ' D'
> 2.33 1.01
> 1.28 1.00
> 1.00 1.33
>
> So cut not only avoids transpose, but is faster (this
> actually surprises me).
>
> In any case, I suspect you wanted to avoid |: not for
> performance reasons, but because it feels too "forced". At
> least, I've often wanted a "natural" way to access different
> axes of arrays (similar to the way that f/ table appears to
> act on the columns of the table). If that's your question:
> I believe transpose and cut are the best we've got.
>
> Gosi wrote:
> > {. I.
>
> This won't work. You still need the transpose (i.e. you need
> [EMAIL PROTECTED]@|: to get the desired result). Also, since [EMAIL
> PROTECTED]
> isn't optimized (yet), you're doing extra work for no benefit.
>
> -Dan
> ----------------------------------------------------------------------
> For information about J forums see
> http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
