Whoops. I forgot it's got a left argument.
Try
x: +/ 12000 sumq >: i. 12000 NB. !
I'd spoilt the spoiler...
Your "f" is a very nice method, but doesn't vectorise easily.
I tried applying this more tacit version:
fa =: (],[-~(]*((<.@%~ maximum &+)~)))/
ab =. fa ab
for pairs of a,b
It's slightly faster, and you can use it with a
power, eg
fa/ ^: number ab
This is fairly fast, but you need to know the value for number!
You can add a counter to fa and make it conditional on 2 < {:
but performance gets poor again.
Mike
On 11/10/2011 9:55 PM, Boyko Bantchev wrote:
> On 11 October 2011 18:29, Mike Day<[email protected]> wrote:
>> ...............
>> Run
>> x: +/ sumq>: i. 12000 NB. very nearly a spoiler
> Mike,
> I haven't read your solution in detail, but was curious to try it, and got
> the following:
>
> x: +/ sumq>: i. 12000
> |domain error: sumq
> | x:+/ sumq>:i.12000
>
>
> On ......... I wrote:
>> f 12000
>> .................
>> ...... very slow for 12000: almost 2 min on my computer.
> I implemented the same algorithm on several other interpreted
> languages and ran it on the same machine as my J solution.
> All were significantly faster than J at 12000:
> Rebol – 19s,
> CLisp – 16s,
> PostScript – 9s,
> Lua, Python – 6.5s,
> Ruby – 3s,
> Scheme (Racket) – less than 0.5s!
>
> J is often amazingly fast at whole-array (non-iterative) operations,
> but seems to be rather unfavourable to iterative (including recursive)
> ones.
> ----------------------------------------------------------------------
> For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
