Evolution Strategies
Single Turnament Considerations

Consideration on the single turnament

Note: The final scores changes at every turnament because the standard strategy that plays in a random way and modifis also the score of the opponents.
In any case, even if the exact scores will be different, the following considerations could be valid in general.

Let's recap the strategies in the game (note: if you modified the parameters in the script files, this list will be automatically updated)

Let's check the score table:

A:BCooperate (1)Defect (0)
Cooperate (1)3 points0 points
Defect (0)5 points1 points

If you did not do it altready, open the tunament in a separate pop up window

"Bad" Strategies

The "Bad" Strategies are indicated in red in the table, these are the strategies that will "defect" also if the opponent will cooperate in the previous round. In thir "DNA" (see program explanation) they have a gene that will determins their colour, the very bad strategies that often defects are red, while the ones that sometime defect are orange.
In the tournament with seven standard participants, in the final classification is evident that the worst strategy is "Always Defect" that usually obtains the lowest score. Also other bad strategies as "Tit for Tat - Trd" and "Tit for Tat - Rim" present low scores.
"Always Defect" always betrays, but for this reason it is discovered fromthe others that looses points on the first round (5 to 0 for "Always Defect"), but they start immediately their revence obtaining 1 point each [in the turnament window, see as an example the game A=1 versus B=6].
"Tit for Tat - Trd", betrays the opponent every 5 rounds. Whtout considerint what the other played, evry 5 times this strategy will defect.
So [see turnament A=1 versus B=3] for the first 5 rounds both cooperates getting 3 points each, the fifth round, "Tit for Tat - Trd" obtains 5 points leaving 0 to the opponent, but the next round the opponent (see in the example A=1 and B=3 a pure "Tit for Tat") get his revenge and obtain 5 points.
In two rounds (the 5th and the 6th) the opponents gets 5 points (0-5 and 5-0) each, versus the 6 (3-3 e 3-3) thye could obtain by cooperating.
"Tit for Tat - Rim" betrays every 5 times as the previous strategy, but if he betrays first, in the next roun forgives the opponent by cooperating. In this way the strategy stops the revenges getting an higher average score than the similar strategy "Tit for Tat - Trd".
"Random" is in the "bad" strategies list because does not care about the opponent behaviour. In the book written by Dawkins this strategy is indicated as the worst, but in our experiment is very variable, getting scores from the lowest to the second best. In the single turnament this strategy has some possibilities, but in the multiple turnaments soon or later it is not able to survive and it evolves in somethig else.

"Good" Strategies

The "good" strategies are in green color in the table, and are those that always cooperates if the opponent show cooperation in the previous round. In thir "DNA" (see the program explanation) thes estrategies has a gene that determins their color.
In the tournament with seven standard participants, in the final classification the best strategy is often "Tit for two Tat" that usually obtains the highest score. This score sometime is beates only by the similar strategy "Tit for Tat".
"Tit for Tat" start cooperating, but if the opponent betrays. the strategy get an immediate revenge in the next round. The game versus itself or versus other similar strategies, [see turnament A=1 versus B=1 or A=1 versus B=2] is boring because everybody cooperates getting 3 points every time. If someone betrays, [see turnament A=1 versus B=6] the strategy get immediate revenge and if the opponent is clever, it will stop immediately, otherwise both will get only 1 point untill the end.
"Tit for two Tat" Is even more cooperative, this strategy will not get his revenge imediately, but it waits to be betrayed twice. It seems ingenuous, but it can win most of the times because it can stop the revenges before the others [see tournement A=2 versus B=3], getting an hogher average score per game.
"Always cooperatea" Is maybe too good, cooperating in any case, but often it gets higher scores than the "bad" strategies

Which is the best strategy?

The results of a single strategy depends on the behaviour of the other. In a turnement with only "bad" strategies" a single "good" strategy will probably be destroyed, while in a turnament with only "good" strategies a single bad strategy will have a good game.
You can download the program and try to invent new strategies, or modify the parameters.

What is next?

If you want to understand how the script works to customize the data and the strategies, read the explanations, otherwise next step will be to try the multiple turnament for generations, where the problem get more interesting.

In this page we have used a single turnament, now, by deciding a minimum survival threshold, it is possible to contiue the experiment pfor different strategies generations.
The rule is easy:
The strategy that can not get the 85% (configurable parameter) of the highest score that will be obtained if both opponent will always cooperate, (3 points per round), the strategy extinguish and is replaced by individuals that wil have an evolution of its DNA.
Also the number of generations is configurable, but pay attention to avoid slowing down too mutch your PC.

Back to the main index of this article

Go to the turnament for generations

Back to the library: www.steppa.net