problems.txt

用C编写的部落寻优的优化算法

1 1 12 21
1 1 1 1
0
0
  0
0
0

1


 1 3
30
30 -20 20 0
0
0 0.00001
0
0 0 100
-40000 -40000 -1

--------------
Put the problem parameter set(s) BEFORE this line. See examples below
If you want to run several problems, precise how many
(very first number after the strategy set) >
WARNING.  A multiojective problem has  to be ran alone
( some pb with the run.txt and run_cleaned.txt files)

DATA FORMAT


------
  <strategy set>  Just for test purpose. Will be hard coded, later.
4 integers = 4 strategies,

strategy 1 is for "bad" particles
 strategy 2 for "normal" ones
 strategy 3 for "good" ones
 strategy 4 for "excellent" ones

For most of problems, the set (1 1 12 21) seems quite good
You may also try  (1 1 12 25) (1 1 12 12)
or simply  (23 23 23 23)   or even (25 25 25 25) for some particular cases

For hybrid problems, you may use (21 21 21 21) (independent gaussian distributions)
(9 9 9 9) is also not bad, but needs the hard coded phi parameter
------
<nonunif set>
 4 floats = 4 nonunif coefficients, one for each strategy.

nonunif coeffs are for distributions of possibilities.

  1 => uniform distribution

 If a nonunif coeff is # 1, for the corresponding strategy,
 the distribution is not uniform
 (see rand_in_hypersphere()).

 if nonunif<0 => Gaussian distribution. Standard deviation = -nonunif

nonunif = (1 1 1 1) is often OK
------
<confin_interv> If <0, no interval (boundary) confinement

------
 <circular_hood> If >0, it is the size of a circular neighbourhood. Then tribes and i-groups are not
       taken into account Mainly useful for comparison.
       For example, by setting it at 3, using always strategy 9, and a constant swarm size (see below)
       you obtain the classical "constricted PSO".

       If <-1, it means "choose a random i-group whose size is -circular_hood"
	   If =-1, "choose a random i-group whose size is (2/N)*ln(eps)/ln(1-1/N)". N=swarm size

------
 <no_best>  0, normal best neighbour
   1, the "best neighbour" is in fact chosen at random
			2, use pseudo gradient to define the "best" informer
			3


------
<adapt>  If 0, complete adaptive method

		 If >0, it is in fact the constant swarm size used for the non adaptive method

------
<linkreorg> Use (1) or not (0) information link reorganization

------
<nb_pb_max> Number of different problems to solve

------
------
<nb_f> number of objective functions to simultaneously optimize
         (needed for multiobjective optimization)
         if <0, use lexicographical order.

------
<function_code>  (see functions.txt or, to be sure, myfunction.c)

     if <0, it means that fuzzification has to be used. The real function code is then abs(function_code)
------
<dimension >
(use <n> times <xmin  xmax  granularity (0 if continuous)>)*

* you can have several lines like this one, for you can have different granularities
along different dimensions
Be SURE that the sum of the all n values is equal to the total dimension

If granularity <0, the problem is seen as a special binary one.
In that case, be sure xmin>-0.5 and xmax<1.5
------
<constrain> For special constraint cases. Usually 0. See constrain.c

			1 for special list of discrete values on the file discrete.txt
------
<target>  Warning. For  a multiobjective problem
							 "target" is only for the first function
------
<precision>
------
< all components of the solution must be different (1), or not necessary (0)>
------
<print_level>
------
< save >    1 => save swarm on swarm.txt
                >1 => save also some data on trace.txt
                In any case, see run.txt for result
				-1 => save energies on energy.txt
			-2 => save the number of evaluations and the best result on file trace_run.txt
			     (in order to plot the convergence curve)
------

<number of runs>
------
<Max_Eval>  Max number of clock ticks if >0. If <0, the absolute value is the max number of evaluations
----
<Max_Eval_2> Second "max number" if you want a whole sequence
<Max_Eval_delta> Step from Max_Eval to Max_Eval_delta
----
if function_code==68 (binary multimodal)
<number_of_peaks>

 ============ Examples you can copy/paste at the beginning of this file
 DO NOT copy/paste the name nor the comment (first text line(s) )

1) DD-sum (sum=200 with 10 different integers in [1,100])
Best parameters for classical PSO (strategy 9): hood 3, size 9
1 1
10
10 1 100 1
0
200 0.1
1

0 0 20
-10000 -10000 -1


2)   DD-product

3) Parabola (Sphere)
 1 3
30
30 -20 20 0
0
0 0.00001
0
0 0 100
-40000   -40000 -1




4) Rosenbrock
 1 4
30
30 -10 10 0
0
0 0.00001
0
0 0 100
-40000  -40000 -1


5) Alpine
1 5
10
10 -10 10 0
0
0 0.00001
0
0 0 100
-15000 -15000 -1


6) Griewank nD, n>2  (min 0 en (100, ... 100))
 1 6
30
30 -300 300 0
0
0 0.00001
0
0 0 100
-40000 -40000 -1

7) Rastrigin
1 7
30
30 -5.12 5.12 0
0
0 0.0000
0
0 0 50
-20000 -20000 -1

8) Fifty_fifty (non combinatorial)
with fuzzification
1 -8


20
20 1 100 1
0
0 0.01
1
0 0 10
 -1000  -1000 -1

9) Ackley

1 9
30
30 -30 30 0
0
0 0.00001
0
0 0 100
-40000   -40000 -1



 10) Foxholes (with fuzzification)
1 -10
2
2 -50 50 0

0
0 0.998004
0
0 0 10
-40000   -40000 -1



11) Apple trees
1 11
3
3 0 1 0
0
1 0.1
0
0 0 1
-200   -200 -1

12) El Farol problem. 100 Irishmen => best solution = 30 at the pub (function value = 15)
after about 50 evaluations
Note that  if the problem is considered as not combinatorial,
it is more difficult to find a solution

1 12
100
100 0 1 1
0
15 0.000001
0
1 0 1
-40000 -40000 -1

13)    Fermat    (find, for example (3,4,5), (24,32,40), (84,13,85) etc.)
1 13
3

3 2 100 1
0
0 0.01
0
1 0 4
-40000    -40000 -1


14)    Knap_sack (10 different integer numbers in [1,100] whose sum is equal to 100

See also DD-sum
1 14
10
10 1 100 1
0
100 0.01
1
0 0 20
-500  -500 -1

15)

16)

17)

18)   2D Linear_system
1 18
3
3 -3 3 0.001
0

0 0.001
0
1 0 1
-40000  -40000 -1

19)


20)

21)   Magic square (3x3) (different integer numbers of [1,100])
  You can also try 16 (4x4), 25 is still possible ... if you are lucky
1 21
9
9 1 100 1
0
0 0.001
1
0 0 10
-10000 -10000 -1

22)    Model tuning
Note: hard coded in myfunction.c. Currently, the target is in fact 1918.95
for (10.073 1.786)
 1 22
2
1 1 100 0.00
1 1  2  0.00
0
0  0.001
0
0 0 10
-5000   -5000 -1



23)    Master Mind  (solution (1,2,3,4) hard coded in myfunction.c)
  Just for fun: it is very bad. You should use my "ultimate" unbeatable MasterMind program ;-)
1 23
4
4 1 6 1
0
0 0.001
0
1 0 1
-1000    -1000 -1


24) Catalan's conjecture
1 24
4
1 2 30 1
1 2 10 1
1 2 30 1
1 2 10 1

0
0 0.001
0
0 0 10
 -40000  -40000 -1

25) //

26)

27)   //

28)


29) Cognitive harmony
Note: need to have the matrix in matrix.txt

1  29
6
6 0 1 0
0
24 0.001
0
0 0 100
-1000   -1000 -1


30) Non specific TSP (just for test). You should use PSO_for_TSP



You need the file tsp_br17 for this problem
You should find a solution after about 2000 evaluations
Note the granularity set to 1
and also the option all_different

1 30
17
17    0.0 16 1
0
39.0 0.9
1
1 0 5
-10000 -10000 -1

With tsp_4
1 30
4
4    0.0 3 1
0
4 0.9
1
1 0 10
-10 -10 -1

With tsp_6

1 30
6

6    0.0 5 1
0
6 0.9
1
1 0 10
-20 -20 -1

31)  Sum of absolute values. You can use it with granularity = 1 (integer problem)

1 31
5
5    -100 100 1
0
0 0.1
0
0 0 1

-40000 -40000 -1

32) Non specific QAP
(use data file qap_scr12)
1 32
12
12    0 11 1
0
31410.0 0.9
1
0 0 10
-5000 -5000 0

33) Multiobjective Lis-Eiben 1
Each run gives a point  whose (f1,f2) is near of the Pareto front
See result in run.txt. (f1,f2) are in the Error columns
2 33
2
2  0 1 0
0
0 0.001
0
0 0 200
-20  -20 -1

34) Multiobjective Schaffer
2 34
2
2  -5 10 0
0
0 0.001
0
0 0 200
-20  -20 -1


35)    Multiobjective F1
 2 35
2
2  0 1 0
0
0 0.001
0
0 0 200

-20   -20 -1

36)      Multiobjective F2

2 35
2
2  0 1 0
0
0 0.001
0
0 0 200
-20   -20 -1

37)    Multiobjective F3        (simple front)


38)  Multiobjective F4          (convex, concave)

39)   Multiobjective F5    (5 Pareto fronts)
2 39
2
2  0 1 0
0
0 0.001
0
0 0 300
-100   -100 -1

40)    Multiobjective F6

41)  Multiobjective Coello F3

42)



43)

44) Jeannet-Messine (cf ROADEF 2003 proceedings p 273)
  min -112.5   in position (3,-7.5, 10)
  The authors define 5 different methods, and
their best result is: 3271 evaluations
1 44
3
1 0 5 1
1 -15 25 0
1 3 10 0
0
-112.5 0.000001
0
0 0 10
-2000  -2000 -1


45)   MINLP X. Yan
 2 45
72
36 0 1 1
12 0 600 0
12 0 400 0
12 0 200 0

0
0 0.01
0
1 0 2
-100000  -100000 -1



46) Tripod (Louis Gacogne)
on [-100, 100]^2, min 0 at (0,-50)
Seems simple but it is in fact difficult
to obtain a success rate > 50%
Strategies 1 1 12 21 => success rate=52%
Strategies 1 1 1 1 => 64%

1 46

2
2 -100 100 0
0
0 0.00001
0
0  0 100
-40000  -40000 -1



 (Compare with strategies 1 1 1 1)

 47) Tripod, with circular constraint, and multiobjective optimization
2 47
2
2 -100 100 0
0
0 0.001
0
0  0 10
-3000 -3000 -1

 48) DeJong f4
1 48
30
30 -20 20 0
0
0 0.0000
0
0 0 50
-20000  -20000 -1

49) Pressure vessel
1 49
4

1 1.125 12.5 0.0625
1 0.625 12.5 0.0625
2 0.01 240 0
0
0 0.0
0
0 0 100
-51818  -51818 -1


50) Coil compressing spring

   Note the constrain=1 (see constrain.c) for discrete values list
1 50
3
1 1.0 70.0 1
1 0.6 3.0 0.0
1 0.2 0.5 0.0
1
0 0.0
0
0 0 100
-12500  -12500 -1

51) Gear train

1 51
4
4 12 60 1

0
0 0.0
0
0 0 100
-19800    -19800 -1

52) Pressure vessel as multiobjective
4 52
4
1 1.125 12.5 0.0625
1 0.625 12.5 0.0625
2 0.01 240 0

0
0 0.0
0
0 0 100
-51818    -51818 -1


53) Coil compressing spring as multiobjective
   Note the constrain=1 (see constrain.c) for discrete values list
6 53

3
1 1.0 70.0 1
1 0.6 3.0 0.0
1 0.2 0.5 0.0
1
0 0.0
0
0 0 100
-12500 -12500 -1



54)


55) Neural Network Training. XOR
1 55
9
9 -30 30 0
0

0 0.001

0
0 0 30
-40000   -40000 -1

56)Neural Network Training. 4 bit parity
1 56
25
25 -100 100 0
0
0 0.001
0
0 0 30
-40000   -40000 0

57) Neural Network Training. Three Color Cube
1 57
46
46 -100 100 0
0
0 0.001
0
0 0 30
-40000 -40000 -1

58) Neural Network Training. Diabetes in Pima Indians
1 58
64
64 -100 100 0
0

0 0.001
0
0 0 30
-40000  -40000 -1

59) Neural Network Training. Sin Times Sin
1 59
26
26 -100 100 0
0
0 0.001
0
0 0 30
-40000 -40000 -1

60) Neural Network Training. Time Servomechanism
1 60
28
28 -100 100 0
0
0 0.001
0
0 0 30
-40000 -40000 -1

61) Move Peaks
	Warning. Some parameters are in fact hard coded (see movpeaks_mc.c)
	Also, the interval of values must be the same for all dimensions
1 61
5
5 0 100 0
0
0 0.00001
0
0 0 50
-500000  -500000 -1

62) Goldberg�s order 3 deceptive problem
	For such a problem, strategy set 1 12 12 21 may be better
	Note the granularity set to -1 (special biary problem)

1 62
60
60 0 1.4999 -1
0
0 0.00001
0
0 0 10
-100000 -100000 -1

63) M�hlenbein's order 5,
   15 binary string
   seen as a 1D problem
1 12 21 21
1 1 1 1
0
0
  0
0
0

1

1 63
1
1 0 32767 1
0
0 0.01
0
0 0 1
-40000 -40000 -1

64) Schaffer's f6
1 1 12 21
1 1 1 1
0
0
  0
0
0

1

1 64
2
2 -100 100 0
0
0 0.00
0
0 0 1
-40000  -40000 -1

68) Binary. Multimodal (D=100, 20 peaks)
 1 68
100
100 0 1 1
0
0 0.000
0
0 0 100
-1000 -20000 -1000
20

 =======================================  to test
 use data file tsp_x50.1      (solution <= 5.89)
21 21 21 21
1 1 1 1

0
0
 0 1
30
50
50    0.0 49.0 1
0
0 0.01
1
1 0 2
-1000000 -1000000 -1
=================== current tests
abs(sin(x) , abs(x)
9 9 9 9

1 1 1 1
3
0
 20 2
99
1
1    1 10 0
0
0 0.00001
0
1 0 2
-1000 -1000 -1