[Jprogramming] Partitions

[Raul Miller]

I suspect not.. (but take my timings with a grain of salt - or better
yet, do them yourself - I was working with a modified version of S
which is quite a bit slower, actually).

But I expect that (except for the edge cases where most of the
instances of a are used) most of the time in S is spent in incremental
updates of a. There are a lot of them.

Thanks,

-- 
Raul

On Mon, Oct 30, 2017 at 10:56 AM, Erling Hellenäs
<erling at erlinghellenas.se> wrote:
> I added the display step. I also modified the index in a, changed them from
> y to y-1 since J has zero as index origin.
>
> I guess 99% of the time is spent in the r=: r,a step. /Erling
>
>
> Den 2017-10-30 kl. 15:43, skrev Raul Miller:
>>
>> Hmm... you're right, that deals with some cases quite a bit better than
>> mine.
>>
>> I can dip into most of them like this:
>>
>> parRDM=:4 :0
>>    select. x
>>      case.1 do.
>>        ,.< i.y
>>      case.2 do.
>>        ((y#2)#:"1 0 }.i.2^y-1) </."1 i.y
>>      case.y do.
>>        ,:<"0 i.y
>>      case.do.
>>        x (1+y-x) P y
>>     end.
>> )
>>
>> ... keeping the definition of P (and combine) from
>> http://jsoftware.com/pipermail/programming/2017-October/049423.html
>>
>> That said, parRuskey still is more efficient for the case where x =
>> y-2 and and y>3 and that's worth some thought.
>>
>> Generally speaking, parRuskey has an advantage of working with
>> partition id rather than partition contents, until the very last step.
>> That's a clean performance advantage (no boxed arrays for most of the
>> work). But ... still needs thought.
>>
>> Thanks,
>>
>
> ----------------------------------------------------------------------
> For information about J forums see http://www.jsoftware.com/forums.htm