Results of Breeding

Changing some parameters of the base configuration, and running sessions on that configurations, led to the following results.

Meaning of the columns in the following table:

1. Configuration
2. Session ID
3. Qualification
4 - 11. Generation of the session

The configurations mean:
base: The base configuration
crossoverProb05: Set crossover probability to 5%
crossoverProb09: Set crossover probability to 9%
decMutationProb10: Decreased mutation probablity to 10%
evaluateElite: Use elite selection strategy
grammarRuleFourRules: Generate always 4 rules
grammarRuleThreeRules: Generte always 3 rules
grammarRuleTwoRules: Generate akways 2 rules
incMaxWrapTo5: Increase max wrap to 5
incMutationProb10: Increase mutation probability to 10%
maxDepth15: Set max depth to 15
nodal: Use a nodal mutation strategy
nodalMaxDerivationTreeDepth1: Use a nodal mutation strategy with derivation tree depth of 1
nodalMaxDerivationTreeDepth10: Use a nodal mutation strategy with derivation tree depth of 10
nodalMaxDerivationTreeDepth2: Use a nodal mutation strategy with derivation tree depth of 2
populationSize40: Popuöation size 40

base	20120619-194512	92,3 31,5 4,1 3,6	1	10	25	50	100	200	500
	20120612-210623	97,3 46,8 6,5 5,8	1	10	25	50	100	200	500
	20120615-192525	29,9 29,9 7,0 7,0	1	10	25	50	100	200	500	1000	3000
	20120613-104221	47,3 24,4 8,9 7,0	1	10	25	50	100	200
	20120615-210453	48,4 19,4 12,3 8,9	1	10	25	50	100	200	500	1000	3000
	20120611-102707	39,7 18,3 11,9 8,9	1	10	25	50	100	200	500	1000
	20120611-184956	45,0 23,5 18,7 9,6	1	10	25	50	100	200	500
	20120613-185311	93,7 47,5 13,9 9,8	1	10	25	50	100	200	500	1000
	20120612-093946	48,2 19,4 13,9 9,8	1	10	25	50	100	200	500
	20120610-233627	32,7 19,3 12,2 10,7	1	10	25	50	100	200	500
	20120614-103322	32,5 14,1 11,0 11,0	1	10	25	50	100	200
	20120614-195730	23,3 18,3 - 11,9	1	10	25	50	100
	20120611-190036	395,5 24,4 16,3 12,1	1	10	25	50	100	200	500	1000	3000
	20120610-094257	90,4 23,5 13,8 12,2	1	10	25	50	100	200	500
	20120609-102738	89,9 32,6 12,2 12,2	1	10	25	50	100	200	500	1000
	20120615-174629	226,5 38,7 16,2 12,3	1	10	25	50	100	200	500	1000
	20120609-171239	276,8 30,9 15,6 13,5	1	10	25	50	100	200	500
	20120608-233350	4446,2 97,3 15,9 13,7	1	10	25	50	100	200	500	1000
	20120610-002640	47,8 24,1 19,4 13,9	1	10	25	50	100	200	500	1000
	20120610-175412	248,6 32,6 14,0 14,0	1	10	25	50	100	200
	20120609-234659	23,8 16,1 - 16,1	1	10	25	50	100
	20120608-192518	1195,5 46,5 19,4 16,2	1	10	25	50	100	200
	20120613-084231	97,2 76,4 29,7 28,9	1	10	25	50	100	200	500
crossoverProb05	20120717-112701	15,8 12,1 8,2 7,0	1	10	25	50	100	200	500	1000
	20120713-104221	30,5 30,2 15,7 13,5	1	10	25	50	100	200	500	1000
crossoverProb09	20120713-104225	191,6 23,6 10,9 8,9	1	10	25	50	100	200	500	1000
	20120717-112706	987,1 91,4 18,6 15,3	1	10	25	50	100	200	500	1000
decMutationProb10	20120615-210453	2340,6 807,8 8,9 5,2	1	10	25	50	100	200	500	1000	3000
	20120615-174629	236,9 14,0 9,0 6,6	1	10	25	50	100	200	500	1000
	20120615-192525	32,3 16,1 9,8 7,5	1	10	25	50	100	200	500	1000	3000
	20120705-222411	1585,1 42,0 18,9 7,5	1	10	25	50	100	200	500	1000
	20120703-221154	47,9 16,2 12,2 7,6	1	10	25	50	100	200	500	1000
	20120702-081329	24,2 19,6 12,3 9,0	1	10	25	50	100	200	500	1000
	20120614-195730	97,2 23,1 - 9,8	1	10	25	50	100
	20120613-185311	46,5 24,2 13,9 9,8	1	10	25	50	100	200	500	1000
	20120705-232615	216,7 69,1 14,0 10,9	1	10	25	50	100	200	500	1000
	20120614-195040	31,3 23,2 13,6 12,0	1	10	25	50	100	200
	20120620-035406	45,4 22,9 15,8 13,6	1	10	25	50	100	200	500	1000
	20120630-151354	75,2 75,2 18,6 13,6	1	10	25	50	100	200	500	1000
	20120702-191056	32,6 24,4 18,6 13,8	1	10	25	50	100	200	500	1000
	20120703-200323	46,9 24,1 16,5 14,1	1	10	25	50	100	200	500	1000
	20120614-103322	1595,1 42,5 23,3 18,4	1	10	25	50	100	200
	20120619-194512	49,0 43,9 18,8 18,8	1	10	25	50	100	200	500
evaluateElite	20120713-104244	23,7 13,8 9,8 9,8	1	10	25	50	100	200	500	1000
	20120717-112722	43,1 10,9 10,9 10,9	1	10	25	50	100	200	500	1000
grammarRuleFourRules	20120721-100435	24,0 18,9 12,3 7,5	1	10	25	50	100	200	500	1000	3000
grammarRuleThreeRules	20120725-110357	42,6 39,5 4,0 2,7	1	10	25	50	100	200	500	1000	3000
	20120721-100429	617,1 19,5 12,3 7,1	1	10	25	50	100	200	500	1000	3000
grammarRuleTwoRules	20120721-100424	16,3 14,0 8,2 6,1	1	10	25	50	100	200	500	1000	3000
incMaxWrapTo5	20120703-221154	30,0 16,4 2,7 2,6	1	10	25	50	100	200	500	1000
	20120630-151354	94,7 31,1 7,1 5,2	1	10	25	50	100	200	500	1000
	20120702-081329	95,5 10,9 9,7 7,0	1	10	25	50	100	200	500	1000
	20120614-180247	49,1 19,7 7,1 7,1	1	10	25	50	100	200	500	1000	3000
	20120705-232615	191,7 68,5 18,7 7,6	1	10	25	50	100	200	500	1000
	20120620-035406	75,3 65,4 16,3 9,0	1	10	25	50	100	200	500	1000
	20120702-191056	32,1 24,1 15,9 10,7	1	10	25	50	100	200	500	1000
	20120705-222411	696,9 22,7 18,4 13,5	1	10	25	50	100	200	500	1000
	20120703-200323	41,3 23,2 15,6 15,6	1	10	25	50	100	200	500
incMutationProb10	20120614-103322	44,1 22,8 3,8 3,3	1	10	25	50	100	200	500	1000	3000
	20120610-094236	31,4 29,6 4,9 3,9	1	10	25	50	100	200	500
	20120614-122812	343,1 32,8 4,7 4,7	1	10	25	50	100	200	500	1000	3000
	20120612-210623	92,3 39,8 12,1 6,2	1	10	25	50	100	200	500	1000
	20120620-035406	32,1 19,6 12,2 6,2	1	10	25	50	100	200	500	1000	3000
	20120611-184956	24,7 16,4 7,0 6,5	1	10	25	50	100	200	500
	20120611-102741	18,3 18,3 10,9 7,0	1	10	25	50	100	200	500	1000
	20120612-093946	698,3 19,6 9,0 7,1	1	10	25	50	100	200	500
	20120613-185311	1069,1 24,2 12,2 9,8	1	10	25	50	100	200	500	1000
	20120611-190036	29,8 18,6 13,9 12,1	1	10	25	50	100	200	500	1000	3000
	20120613-104221	31,8 16,1 14,0 12,1	1	10	25	50	100	200	500
	20120613-084231	72,0 45,3 13,8 12,2	1	10	25	50	100	200	500	1000
	20120610-234209	4481,0 44,0 19,3 13,8	1	10	25	50	100	200	500	1000
	20120610-175427	80,1 23,1 14,1 14,0	1	10	25	50	100	200
	20120611-113733	80,4 77,9 22,8 18,5	1	10	25	50	100	200	500
maxDepth15	20120717-112732	97,4 48,9 6,2 5,8	1	10	25	50	100	200	500	1000
	20120713-104245	40,6 22,6 15,6 13,4	1	10	25	50	100	200	500	1000
nodal	20120824-163537	221,9 193,6 177,1 152,9	1	10	25	50	100	200	500	1000	3000
	20120824-163449	392,5 310,1 190,0 171,6	1	10	25	50	100	200	500	1000	3000
	20120827-190453	353,4 295,8 188,7 186,1	1	10	25	50	100	200	500
	20120827-190432	556,5 258,8 199,0 194,2	1	10	25	50	100	200	500
nodalMaxDerivationTreeDepth1	20120809-194653	18,4 13,5 13,4 9,6	1	10	25	50	100	200	500	1000	3000
	20120814-181652	79,3 30,2 15,9 15,7	1	10	25	50	100	200	500	1000
nodalMaxDerivationTreeDepth10	20120809-194705	47,9 14,1 6,1 4,5	1	10	25	50	100	200	500	1000	3000
	20120814-181704	32,2 16,3 12,2 9,0	1	10	25	50	100	200	500	1000
nodalMaxDerivationTreeDepth2	20120809-194658	76,9 13,6 6,6 3,8	1	10	25	50	100	200	500	1000	3000
	20120814-181658	1360,4 48,4 12,3 8,9	1	10	25	50	100	200	500	1000
populationSize40	20120713-104234	46,0 29,9 9,6 4,1	1	10	25	50	100	200	500	1000
	20120717-112713	91,1 47,5 10,9 9,8	1	10	25	50	100	200	500	1000

First Results in Rule Optimizing

Using the geva library from the natural computing research and application group leads to the first results.

Using geva you must invert the fitness value. Low value means high fitness. A value of 9000 means that the opponent is almost immedately thrown out of the arena. So we can see the breeding was really very successful.
This are the rules that where breeded.

RULEBLOCK GENERATED
 OR : MAX ; 
 ACCU : MAX;
 RULE 1 : IF  opponentDir IS invisible THEN leftVelo IS backward , rightVelo IS forward  ;
END_RULEBLOCK

 
The only generated rule focuses only on the opponent, and means turn around as long as you do not see the opponent. Otherwise use the default behaviour, which is go straight ahead if no rule matches. Very simple but effective.

The opponent was the manually desined fuzzy controlled robot using the following rules.

RULEBLOCK No1
    AND : MIN;
    ACT : MIN;
    ACCU : MAX;

    RULE 1 : IF opponentDir IS invisible
                    AND frontBorderDist IS near
                    AND leftBorderDist IS near
                    AND rightBorderDist IS near 
                THEN
                    rightVelo IS fastBackward,
                    leftVelo IS fastforward;

    RULE 3 : IF opponentDir IS invisible
                    AND (leftBorderDist IS medium OR leftBorderDist IS far)
                    AND rightBorderDist IS near
                THEN
                    leftVelo IS standstill,
                    rightVelo IS forward;

    RULE 2 : IF opponentDir IS invisible
                    AND leftBorderDist IS near
                    AND (rightBorderDist IS medium OR rightBorderDist IS far)
                THEN
                    rightVelo IS standstill,
                    leftVelo IS forward;

    RULE 4 : IF opponentDir IS left
                THEN
                    leftVelo IS fastforward,
                    rightVelo IS forward;

    RULE 5 : IF opponentDir IS right
                THEN
                    leftVelo IS forward,
                    rightVelo IS fastforward;

    RULE 6 : IF opponentDir IS center
                THEN
                    leftVelo IS fastforward,
                    rightVelo IS fastforward;

END_RULEBLOCK


You can see that robot, fighing against a clone of itself using the follwing link: Manual vs Manual

Another low variance long running session

This will probably be the last parameter optimizing session, as parameter optimizing does not bring the expected results.

The average fitness never exceeds 1000. After 1000 generations there is an improvement in the maximum values, but the no further advancements.
Have a look at the robots after 0, 300, 1000 and 2216 generations.

A long running low variance breeding session

This breeding session generated the absolute maximum average (and max) fitness. Relatively fast it rises to more than 10000 in average. That means that more than half of the individuals of a population could win their matches against the reference robot.
Why could there no further improvement? No idea.
See the robots in action after:
0, 1000, 2000 and 2384 generations

Low variance initial generation

As the performace of the results of parameter optimizing did never get even near the results of the manually designed fuzzy controllers, I tryed to make the inital generation almost equal to the manually choosen parameters (low variance). The GA sould not find a better solutioion somewhere in the waste area of parameters, but find an optimized solution staring at a almost optimal solution.
Here are the results of that.

The fitness rises immediately to a much higher average and maximum than in the sessions with random initial generation (see the results in older posts).
Here are some of the best performers of their generations. The violet sumo is the 'breeded' one.

• 0 generation
• 1 generation
• 100 generations
• 500 generations
• 800 generations

Another long running breeding session

See the behaviour after several generations.

• 1 generation
• 370 generations
• 1354 generations
• 2168 generations
• 3066 generations
• 4116 generations

A long running breeding session

I let a session run for several days (on my not very fast computer) to see what happens after several thousand generations.

The result of that session shows that a minimum improvement can be achieved after a view hundred generations.
Check the behaviour of the fittest individual in that phase.

• 0 Generations
• 188 Generations
• 1943 Generations 

The pink sumo robot contains the 'breeded' controller whereas the blue one is controlled by a manually designed fuzzy controller. Obviously the breeded robot does not have a chance towin against the manually designed one. Even not after more than 1900 generations.
Here are some more breeded robots after the increase of fitness at about 3500 generations.

• 3863 Generations
• 4089 Generations