From: Philip Rhoades <phil@...>
Date: 2010-10-08T22:37:38+09:00
Subject: Re: A Real World example for Ruby to "compiled" version discussion

People,

I have accumulated the questions from different people into this response:


On 2010-10-07 23:12, Robert Klemme wrote:
> On Thu, Oct 7, 2010 at 2:57 AM, Ryan Davis<ryand-ruby@zenspider.com>
> wrote:
>>
>> On Oct 6, 2010, at 09:55 , Philip Rhoades wrote:
>>
>>> The main problem I see with some of the Ruby conversions that I
>>> have looked at (eg RubyInline) is that the performance problem
>>> comes in in repeating the WHOLE simulation with different
>>> starting parameters which is done thousands of times - so it is
>>> not like you have a single recursive algorithm which is a
>>> bottleneck that can be optimised or rewritten in C or something.
>>> There are lots of little steps that happen millions of times . .
>>
>> Could you expand on this??
>
> Specifically
>
> - What properties (conceptually) does an individual / a population
> have?


This is an Individual Based Model (IBM) as opposed to populations that 
can be modelled as a whole with equations etc - so the simulation 
reflects this.  In the C/C++ version an individual organism is an object 
with a number of attributes: ID, parent ID(s), (microsatellite) allele 
repeat length, randomizer int etc


> - What calculations are done for each individual / population?


The genetics of the population changes as the individuals change because 
of mutations which affect repeat length of microsatellites ie they are 
not selected against/for (neutral mutations).


> - What kind of iteration is applied?


For this particular discussion:

for k in 1..K		# generations
   for x in 1..X		# array width
     for y in 1..Y	# array height
       destroy parent ordered list
       destroy tmp ordered list
       offspring list becomes parent list
       for i in 1..I	# organism in parent ordered list
         asexually reproduce offspring with possible mutation
         possibly migrate offspring to neighbouring array cell
       end
     end
   end
end

data is written to disk after each generation and genetics statistics 
are done later.


> How much data is carried for each individual? a few alleles? a lot?
> the whole human genome?


For this exercise, one loci and allele (asexual reproduction) - multiple 
loci can be modelled by multiple runs.


> What is done in the innermost loop? I imagine you select a pair of
> individuals, cross them, etc to produce more individuals. Or perhaps
> an individual dies or migrated to another population.


Asexual reproduction in this case and see above.


> What do you do to run a new experiment? generate the initial
> population, specify the rules for combination, depth, migration, etc?
> gather data and process it at the end?


A script calls the simulation program hundreds of times with different 
starting populations.  These starting populations are the output of the 
same simulation program which has previously been run, starting with 
different random number seeds, until they reach "quasi-equilibrium" 
(typically 500 generations).  Also see above.


> How much code do you need to write or rewrite for an experiment? How
> many parameters are involved?


The code doesn't change - a configuration file and the starting 
(initial) generation can be changed.


> Can you give some examples of aspects of the C++ program that you
> hope to improve by moving the sim to Ruby?


Not really - it was a while ago now that I did the actual coding (I 
don't need to touch the code much these days) and I just remember it was 
a major pain trying to learn about the eccentricities of OOP with C++ - 
I'm sure it would have been a much more pleasant experience doing it in 
Ruby . .

I actually don't need to change to Ruby now - I just need to write a 
thesis . . What would be nice would be to further develop the program, 
make it more generalised, more friendly for other users, be able to 
respond to requests for enhancements/changes more quickly etc.


> Can you say something about the context of your modeling? How is your
> simulator different from others used in the field?


There are a couple of particular issues we wanted to investigate and my 
supervisor thought that the existing stuff that was around then (about 
nine years ago now) didn't conveniently allow us to do what we wanted - 
maybe it is has changed now but I haven't had much exposure to other 
programs.  Also, a lot of them then were written in some sort of 
graphical environment for Windows or Macs and I wanted to have a CLI 
interface on Linux and have more direct control over what was happening . .

Thanks,

Phil.
-- 
Philip Rhoades

GPO Box 3411
Sydney NSW	2001
Australia
E-mail:  phil@pricom.com.au