[Push] Advice on Fitness

[Rud Merriam]

Hi Lee,

My approach was similar. I modified your approach to do:

	-1000 if nothing on boolean stack
	10 if correct value
	0 if incorrect value
	sum these over the test values

I'm testing values from 0 to 19 as cited in one of the papers. I plateau at
values of 100. Those programs are using various techniques for generating a
constant boolean value. The few times the value is above 100 is with a
BOOLEAN.RAND instruction and those programs get tossed in the next
generation.

Let me describe the overall processing to see if that needs changing. I loop
through:

1. Purge any program that has less than the average fitness of the
population
2. Bring population up to 80% of maximum using crossover if population is
larger than the tournament size
3. Complete the population with random programs
4. Run the program and record the fitness for each program

One question is whether the parents should be retained in the next
generation, which I am doing. 
 
 - 73 - 
Rud Merriam K5RUD 
http://mysticlakesoftware.com/


> -----Original Message-----
> From: Lee Spector [mailto:lspector at hampshire.edu] 
> Sent: Thursday, May 06, 2010 5:24 PM
> To: Rud Merriam
> Cc: push at lists.hampshire.edu
> Subject: Re: [Push] Advice on Fitness
> 
> 
> 
> What are you using now? 
> 
> In my clojure version of the odd problem I use the code below 
> as the error function, which in English says try inputs 0, 1, 
> 2, 3, 4, 5, 6, 7, 8, 9, and for each push the input on the 
> integer stack and also make it available via an "in" 
> instruction (which pushes it on the integer stack -- I do 
> this via an auxiliary stack but that's just an implementation 
> detail), and then run the program and look at the top item of 
> the boolean stack at the end. If there's no item on top of 
> the boolean stack the error for that case is a penalty of 
> 1000. If there is an item then the error for that case is 0 
> if it's right for the input and 1 if it's wrong. The total 
> error ("fitness," for which lower is better) for the 
> individual is the sum of the errors for the 10 cases. I run 
> this with all of the implemented push instructions available, 
> the "in" instruction, and random positive integers less than 100.
> 
>  -Lee
> 
> 
> 	 (fn [program]
> 	   (doall
> 	    (for [input (range 10)]
> 	      (let [state (run-push program
> 				    (push-item input :auxiliary
> 					       (push-item input :integer
> 							  
(make-push-state))))
> 		    top-bool (top-item :boolean state)]
> 		(if (not (= top-bool :no-stack-item))
> 		  (if (= top-bool (odd? input)) 0 1)
> 		  1000)))))
> 
> 
> On May 6, 2010, at 5:27 PM, Rud Merriam wrote:
> 
> > I have my PushX running with a crossover operation and 
> trying to solve 
> > the ODD problem, i.e. Boolean result if an integer is odd or even.
> > 
> > I'm not having any success so I'm guessing my fitness test is not 
> > crafted properly. What would everyone suggest?
> >   
> >  - 73 -
> > Rud Merriam K5RUD
> > http://mysticlakesoftware.com/
> > 
> > _______________________________________________
> > Push mailing list
> > Push at lists.hampshire.edu 
> > https://lists.hampshire.edu/mailman/listinfo/push
> 
> --
> Lee Spector, Professor of Computer Science
> School of Cognitive Science, Hampshire College
> 893 West Street, Amherst, MA 01002-3359
> lspector at hampshire.edu, http://hampshire.edu/lspector/
> Phone: 413-559-5352, Fax: 413-559-5438
> 
> Check out Genetic Programming and Evolvable Machines: 
> http://www.springer.com/10710 - http://gpemjournal.blogspot.com/
> 
>