At the end of my last development-related post, I said I should probably run more tests to confirm the apparent first-player advantage in Beaver. Well, I've done that now.
Without coding any of the minimax stuff, I've run each existing strategy against each other, going both first and second. While the cleverer strategy always has a higher win percentage, regardless of whether it plays first, tests in which both players use the same strategy continue to show a first player bias:
| Strategy | P1 win | P2 win | Draw |
|---|---|---|---|
| Adequate | 46.1% | 38.2% | 15.7% |
| Dumb | 47.4% | 38.0% | 14.6% |
| Dumber | 44.1% | 39.9% | 16.0% |
| Dumbest | 59.5% | 27.9% | 12.6% |
So, the next step would be to create a few more strategies to test it with. :)
At least, that would be the next step if my computer hadn't just died, taking the source code with it (and god knows what else). Thankfully, I've still got a fairly recent copy of the EXE and the game logic assembly on a USB flash drive, so if all else fails, I'll be able to use Lutz Roeder's .NET Reflector to recover most of the source code. Albeit source with poorly named variables. Hey ho. Guess it's time to get a new PC...
No comments:
Post a Comment