CODEWITHSUNDEEP
optimizationparametersscilabestimation
HerClau
HerClau

Reputation: 171

Scilab: Parameters estimation on Lotka Volterra model Scilab

I have tried to reproduce the script of the following link: Parameters estimation on Lotka Volterra model with Scilab

And I get similar errors in the results to those described in it.. You could guide me to run the script without errors. Gracias Hermes

Upvotes: 0

Views: 220

Answers (1)

PTRK
PTRK

Reputation: 891

Solution for Scilab 5.5.1 or smaller

The problem is that the solver somehow reach a point where it cannot solve the ode on every t and stops at a certain point. Thus your y_calc is smaller than y_exp in size.

If this is not a problem for you, change diffmat on line 6 of the Differences function to 

diffmat = y_calc' - y_exp(1:size(y_calc',1),:)

Solution for Scilab 6.0.0 or greater

The ode function now raise an error when it fails to compute in Scilab 6.X.X.

Using the try ... catch ... end statements, we can let the program deal with it.

So a wrong (and i will explain why after) solution is changing line 5 of Differences function to

try // test if ode works normally
  [y_calc,odew,odeiw]=ode(y0',t0,t,list(LotkaVolterra,c,n,m,e)) 
catch // if an error is raised
  y_calc = y_exp' // Put a value when the computations fails
end
diffmat = y_calc' - y_exp(1:size(y_calc',1),:)

You would still get warning from ode.

Why it is bad

Since you asked for a running program, i've used the try catch statement to toss away the error and let the program continue. You should not. It means that your problem is ill defined: the parameters are poorly chosen or it's not numerically solvable. I recommend to work on the mathematics behind the program and find why some parameters may break the ode algorithm.

And there's another reason you should not follow my answer. As you see, when I catch an error, I give to y_calc some value. But that's mathematically absurd because an optimization problem rely on the fact that the solution approximate at best the initial problem. I'm 99% sure that the y_calc = y_exp' statement will gives you unusable results.

Upvotes: 0

Related Questions