The performance bug has now been fixed for the next release.

   y=: 1e6 [EMAIL PROTECTED] 2
   ts=: 6!:2 , 7!:[EMAIL PROTECTED]
   ts '(}.+.}:) y'
0.00510168 3.14662e6
   ts '2+./\y'
0.00363596 2.09843e6
   ts '3+./\y'
0.00973976 1.04966e6
   ts '4+./\y'
0.0114389 1.04966e6
   ts '5+./\y'
0.0122412 1.04966e6
   ts '50+./\y'
0.0124204 1.04973e6



----- Original Message -----
From: Roger Hui <[EMAIL PROTECTED]>
Date: Friday, March 14, 2008 13:55
Subject: Re: [Jprogramming] expanding timed data into time-series
To: Programming forum <[email protected]>

> There is a performance bug somewhere in  2 +./\y .
> 
>    y=: 1e6 [EMAIL PROTECTED] 2
> 
>    ts '2 +./\y'
> 0.124445 2.09856e6
> 
>    ts '3 +./\y'
> 0.024378 1.04966e6
>    ts '4 +./\y'
> 0.024892 1.04966e6
>    ts '5 +./\y'
> 0.0258469 1.04966e6
> 
> 
> 
> ----- Original Message -----
> From: Dan Bron <[EMAIL PROTECTED]>
> Date: Friday, March 14, 2008 11:00
> Subject: Re: [Jprogramming] expanding timed data into time-series
> To: Programming forum <[email protected]>
> 
> > Arie:
> > >  I do not get your result?
> > 
> > Put this down to "the vaguaries of benchmarking".  The 
> > difference is not significant in either case.
> > 
> > OTOH, I just did different test:
> > 
> >            ts 
> > =: 6!:2 , 7!:2@:]
> >             Y =: ?.@:$&2]1e7
> >     
> >            
> > ts&>'(}. +. }:) Y';'2 +./\ Y'
> >     0.0517359 5.03327e7
> >          0.227543 
> 3.35559e7>    
> >            
> > 0.01 round (%"1<./) ts&>'(}. +. }:) Y';'2 +./\ Y' [ load'misc'
> >            1 1.5
> >     4.69   1
> > 
> > This difference *is* significant, and it corroborates Oleg's 
> > assertion:
> > >  I thought it was shown that old style shift
> > >  is more efficient than infix. 
> > 
> > which surprises me, given the links I cited 
> > before.    Roger, can you clarify or 
> > elaborate?  
> > 
> > Oleg, I haven't been watching the Forums closely lately; would 
> > you point me to the messages that highlighted this behavior 
> > (i.e. shift+operate is faster than infix)?
> > 
> > -Dan
> > 
> > PS:  This latter benchmark is not relevant to the problem 
> > at hand, because of the requirement for catentation I 
> indicated 
> > earlier.  Neither method is a clear winner in that case 
> > (i.e. their performance is comparable).
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to