Re: Optimising the Keen-Minsky Model

[richard]

On Sun, May 27, 2012 at 06:28:05PM +0200, c. wrote:

> Here is a modified version of your code, I could get a ~ x2 speed up by 
> refactoring and partial vectorization
> plus a further ~ x2 speedup by running in parallel with parcellfun (I only 
> have 2 cores, if you have more that could be better for you).
>
It will take me some time to get my head around this: I'll learn something from 
your 
code and the speed up will certainly help if I attempt to model the Romans (~ 
50BC -> 450AD)
 
> >> clear all, close all, tic, KeenMinsky02_refactored, toc 
> Anyway it looks like you are solving an ODE system, maybe you should look at 
> LSODE 
> 
> HTH,
> c.
I see a lot to think about. Another objective is to help think about tools to 
work on 
non-linear time varying systems modelled as Y=AX + BU 
hence in the long run your coding methods may be even more beneficial than the 
speedup. I'll follow up after I've run the model a few times. 

> 
> P.S. I noticed the 12th row of X is filled but never used ... maybe that is a 
> bug?
> 
I have a later version that includes depreciation and an interaction between 
wages and prices, hence X12 is used for that, plus a needed new, time-varying 
timelag. 
Trying to change things and at the same time keep the system, and results the 
same
is challenging :-)

A Thousand Thanks ! 

Richard A Lough