Re: [F-Script-talk] Adding Math Methods to Arrays

[Philippe M. <pm...@ac...>]

Great. BTW, if you haven't read it yet, you will find some useful  
information on the design of the optimized  array programming model in  
the OOPAL paper, section 8.2, at http://www.fscript.org/documentation/OOPAL.pdf

Philippe

Le 24 juin 08 à 00:23, Peter A Passaro a écrit :

>
> Excellent, thanks again Philippe, that is exactly what I was looking
> for. I do need every bit of speed I can get as I am working with data
> sets of a few Gb in size and performing a number of intensive
> operations on them, so I'll try to apply my optimization tricks where
> appropriate. I'll start experimenting and let you know how it turns  
> out.
>
> Peter
>
> On 23 Jun 2008, at 20:23, Philippe Mougin wrote:
>
>> Peter,
>>
>> You'll not need to modify the existing F-Script source code. The
>> system is extensible and you can achieve what you want in a modular
>> way by putting your code in Objective-C categories that you can
>> build as separate bundles and load dynamically if you want.
>>
>> For method that deal with whole arrays (i.e., that have no meaning
>> for single numbers; e.g. calculating an average value given an array
>> of numbers), you should first implement your method in a category of
>> NSArray, in order to deal with the situation where you'd have an
>> array of NSNumber objects. See for instance how the method "reverse"
>> is implemented in F-Script's FSNSArray category. In your case you'll
>> also need to make sure you raise an error if objects other that
>> NSNumbers are found in the array. Then, if you want to have
>> optimized implementations for situations where F-Script is able to
>> determine that the array is made of numbers, you should add little
>> stubs in a category of Array (again, for an example, see how the
>> "reverse" method is implemented in the Array class) as well as your
>> optimized implementations in a category of ArrayRepDouble (see how
>> the "reverse" method is implemented in the ArrayRepDouble class).
>>
>> For math methods that make sense on single numbers (like cos, sin,
>> abs, etc.), you should first implement your method on single number
>> objects. To do that, implement them for NSNumber objects in your own
>> category of NSNumber (look at FSNSNumber for examples). Note that,
>> at this point, you automatically get array programming for these
>> methods. F-Script will take care of automatically broadcasting
>> messages when you deal with arrays. In general this is fast enough.
>> If you need to go even faster, you can implement optimized versions
>> of your methods in a category of ArrayRepDouble. You'll just have to
>> add simpleLoop_xxx or doubleLoop_xxx methods in your category
>> (depending on the message pattern your math operation supports). See
>> ArrayRepDouble.m for example of implementing these simpleLoop_xxx
>> and doubleLoop_xxx methods.
>>
>> Feel free to ask for further clarifications if needed.
>>
>> Philippe
>>
>> Le 23 juin 08 à 19:29, Peter A Passaro a écrit :
>>
>>>
>>> Hi Philippe and All,
>>>
>>> I'm starting to work my way through the F-Script framework code, and
>>> the first thing I would like to do is add some additional math
>>> operators to the ArrayRepDouble class. Could you possibly give me a
>>> quick overview of the framework classes I will need to modify in
>>> order
>>> for this to work properly?
>>>
>>> Specifically, what I would like to add is a bunch of statistics and
>>> vector operations for arrays of doubles. I have been writing Obj-C
>>> wrappers (or alternative functions) for many of the functions in the
>>> GNU Scientific Library, and I would like to bring these into F- 
>>> Script
>>> so I can avoid the back and forth conversion of arrays of doubles-
>>>> NSNumbers->doubles.
>>>
>>> I may also be interested in contributing a plotting/graphing
>>> application for arrays of doubles to F-Script, I have one I am
>>> developing for visualizing large time series data sets that will
>>> probably work well. Let me know if this might be interesting.
>>>
>>> Best,
>>> Peter
>>>
>>> Peter Passaro
>>> Research Officer
>>> Dept of Informatics/Dept of Biology
>>> Pevensey III
>>> University of Sussex
>>> Falmer, Brighton BN1 9QH, UK
>>> Office + 44 (0)1273 877590
>>> P.A...@su...