Genetic Algorithm Questions:

1. What is a good population size ?
	Why?
	Is this true for all problems?
	In the absence of needed information what
	value should we use ?

2. What is a good crossover probability ?
	Why?
	Is this true for all problems?
	In the absence of needed information what
	value should we use ?

3. What is a good mutation probability ?
	Why?
	Is this true for all problems?
	In the absence of needed information what
	value should we use ?

   What is a good value of ...?
	Why?
	Is this true for all problems ?
	In the absence of needed information what
	value should we use ?

4. Why ask why?

5. How the heck do I found out the answers?
	Experiment.
	


Write a shell script/program to 

	1. Take the output performance measures
(max, avg, min, ...) over many runs of the
genetic algorithm (a run has identical GA
parameter settings) where each run was obtained
with a different random seed and calculate the
average of the maximum fitness of the first
generation results for each run of the genetic
algorithm.

ofile_1	ofile_2	ofile_3	ofile_4	ofile_5 ....

each ofile has a rows of numbers corresponding
to the performance measures for that generation/row.

row1 corresponds to generation 1
row2 corresponds to generation 2
row3 corresponds to generation 3
...

Take the first row from ofile_1, ofile_2, ofile_3,....ofile_n

each row contains
gen	max	avg	min	...

let max_i be max fitness value in ofile_i



calculate

	max_avg = (max_1 + max_2 + ... + max_n)/n

	where n is the number of output files.

	This is our average of the maximum fitness
for the generation corresponding to the row we
extracted. This maxavg should better represent the
max fitness value in the generation under
consideration.

We do this for every row and every measured
performance variable of interest.

	maxavg	average of the maximum fitness
	avgavg	average of the average fitness
	minavg	average of the minimum fitness

	...

AND THIS IS WHAT I WANT TO SEE PLOTTED ON THE GRAPHS.

	X-Axis		generations

	Y-Axis		maxavg, avgavg



More Questions:

1. What is a good encoding?
	We are using a simple encoding

	dn....d3 d2 d1 d0 =
dn * 2^n + ... + d3 * 2^3 + d2  * 2^2 + d1 * 2^1 + d0 * 2^0

	What about gray codes? 
	What encoding should we use?
2.
Corvin asked
	Should we try to ensure that each bit/gene has  
	about the same effect ?
	Example: changing d3 	changes the value of
	the parameter by 2^3 while changing d0
	changes the value of the parameter by 2^0 = 1

Answer: Usually we don't know enough to ensure that
	each gene's contribution is the same.



How do we attack a problem like 

maximize f(x) = x^2 in the range

	-5.12 to 5.11


and 

	minimize

	f(x) = x^2  in the range

	-5.12 to 5.11

and

	minimize

	f(x1, x2, x3, x4, x5) = x1^2 + x2^2 + x3^2 + x4^2 + x5^2

in the range 	-1.024 ..  1.023  for x1
		-5.11  ..  5.11	  for x2
		-3.45  ..  6.89   for x3

		...




Page 108 in textbook (available in the bookstore now)

Try	F1 thru F4