Remember that
name=: c}x,:name
, with c boolean, runs fast and in-place. If you care about speed, time
the version ith assignment to the same name as one of the operands.
Henry Rich
Fraser Jackson wrote:
> The best fill code depends on the size of the data arrays and the density of
> missing elements but after some experimentation over 10<n<1e6 and missing
> percentages from 50 down to 1 the two simple steps
>
> m =: I. a: = data
> (<"0 m{fill) m}data
>
> seem to perform faster than Dan's
> (data=a:)}data ,: <"0 fill
>
> and by a factor of at least 2 for the other suggestions.
>
> In many problems the appropriate missing value is only calculated for
> missing cases simplifying the expression even further.
>
> Fraser
> ----- Original Message -----
> From: "Raul Miller" <[email protected]>
> To: "Programming forum" <[email protected]>
> Sent: Friday, August 14, 2009 1:14 AM
> Subject: Re: [Jprogramming] filling in missing values
>
>
> On Thu, Aug 13, 2009 at 7:49 AM, John
> Randall<[email protected]> wrote:
>> Dan Bron wrote:
>>>> Suppose I have some data, and some default values to
>>>> use in case there is no data:
>>>> data =. a:,(1;2;3),a:,(3;2)
>>>> fill =. 10 20 30 40 50 60 70
>>>> The result I want is:
>>>> 10;1;2;3;50;3;2
>>> I prefer the mask}data,:fill idiom:
>>>
>>> (10;1;2;3;50;3;2) -: (data=a:)}data ,: <"0 fill
> .....
>> And I like the selection vector approach: sv {"_1 x,.y .
>>
>> (10;1;2;3;50;3;2)-:(data=a:) {"_1 data,.<"0 fill
> .....
>
> And I prefer the arithmetic approach
>
> (10;1;2;3;50;3;2) -: <"0 ({.&> data) + fill * data=a:
> 1
>
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
