search

schrum2 / MM-NEATv2

MM-NEAT version 2.0 is no longer supported. Please get MM-NEAT 3+ from https://github.com/schrum2/MM-NEAT
Other
11 stars 5 forks source link

Number of games against static opponent fitness/other score #444

Closed schrum2 closed 7 years ago

schrum2 commented 7 years ago

While re-reading the Othello coevolution paper (http://citeseerx.ist.psu.edu/viewdoc/download;jsessionid=10AB4B0966FEE51BE133255498065C42?doi=10.1.1.580.8400&rep=rep1&type=pdf) I noticed that they evaluated performance against the WPC by playing 5000 (!) games against it with randomized moves. That's a lot. We shouldn't do that much, but we probably should have a parameter that determines how many games are played against the static opponent. Currently, only one game is played ... the parameter can be used to play multiple games and average the win/lose/draw fitness (or whatever score is being tracked)

DarwinJohnson commented 7 years ago

Created the Static Opponent Trials Parameter called boardGameStaticOpponentRuns. The agents will face the Static Opponent a number of times equal to this Parameter, and the Scores from the runs will be averaged. This Parameter is defaulted to 1, so no previous Batch Files will be affected. Ran an Othello Batch File to see if this Score was being tracked correctly, and it appears to be working.

schrum2 commented 7 years ago

One more small request before closing this: Make the StaticOpponentFitness return a score that is the percentage of wins (using the new fitness from #445 ) instead of the WinLoseDraw average.

DarwinJohnson commented 7 years ago

Now the Static Opponent Fitness track the Win Percentage instead of the Simple Win-Lose-Draw Fitness.