These were run on the same laptop that I used in the contest in 2010, which as I recall is slower than the contest virtual machines were. Bots which need the time may play weaker in my tests than they did in the contest. There are some.
Some of the opponent names here do not match the correct names on the official site. Sorry.
300 games each against earlier versions on random maps. As always, this is not good for judging strength against other opponents. I like these comparisons anyway because they give a feeling of progress.
| bot | win rate | wins | losses | draws |
| oddshrimp3.2 | 56.0% | 114 | 78 | 108 |
| oddshrimp3.1 | 57.5% | 117 | 72 | 111 |
| oddshrimp2.4 | 60.7% | 168 | 104 | 28 |
| oddshrimp2.3 | 67.7% | 193 | 87 | 20 |
| oddshrimp2.2 | 66.7% | 186 | 86 | 28 |
| oddshrimp2.1 | 77.7% | 228 | 62 | 10 |
| oddshrimp14 | 79.8% | 238 | 59 | 3 |
| oddshrimp12 | 85.5% | 255 | 42 | 3 |
The big testing news is that, after several rounds of confusion, I finally got #1 bocsimacko working. I tested its strength and got a rating a hundred points lower than the official contest rating (meaning it’s still #1 by another hundred points). Running on my slow laptop may hurt it.
200 games each against 20 opponents on random maps, 4000 games total. The best estimate elo works out to the low 3440’s, corresponding to rank 15 or thereabouts, halfway between davidjliu and wagstaff on this list. As always, the uncertainty is somewhat large even with 4000 games because there are only 20 opponents, which may not be entirely representative.
This time I divided games into “long”, lasting the full 200 moves and decided on ship count, and “short”, ending earlier when one side is annihilated. I broke out winning rates for short and long games and mean length of short wins and short losses. So for example against bocsimacko, 91% of games were short and few of those were wins, but over half the long games were wins. Among the short games, winning took 102 turns on average and losing took 82 turns, reflecting that bocsimacko defends well and wins efficiently.
| opponent | rank | win rate | wins | losses | draws | short games | short win rate | short win turns | short loss turns | long win rate |
| bocsimacko | 1 | 23.5% | 44 | 150 | 6 | 91.0% | 20.3% | 102 | 82 | 55.6% |
| iori | 2 | 35.0% | 62 | 122 | 16 | 76.5% | 30.1% | 106 | 91 | 51.1% |
| GreenTea | 8 | 42.5% | 79 | 109 | 12 | 86.5% | 41.6% | 90 | 99 | 48.1% |
| dmj111 | 12 | 43.8% | 83 | 108 | 9 | 86.5% | 44.5% | 81 | 104 | 38.9% |
| davidjliu | 13 | 51.5% | 97 | 91 | 12 | 80.0% | 50.6% | 93 | 87 | 55.0% |
| wagstaff | 17 | 51.7% | 99 | 92 | 9 | 85.0% | 54.1% | 95 | 116 | 38.3% |
| medrimonia | 18 | 53.5% | 103 | 89 | 8 | 57.0% | 69.3% | 85 | 109 | 32.6% |
| smloh | 19 | 69.0% | 136 | 60 | 4 | 87.0% | 75.9% | 79 | 115 | 23.1% |
| Neverstu | 28 | 58.2% | 113 | 80 | 7 | 84.5% | 59.2% | 84 | 123 | 53.2% |
| Manwe | 31 | 56.5% | 108 | 82 | 10 | 82.5% | 56.4% | 87 | 106 | 57.1% |
| animatroid | 36 | 67.0% | 131 | 63 | 6 | 86.0% | 71.5% | 83 | 102 | 39.3% |
| mogron | 46 | 68.5% | 136 | 62 | 2 | 77.5% | 78.1% | 84 | 95 | 35.6% |
| deccan | 47 | 74.0% | 147 | 51 | 2 | 81.0% | 77.2% | 80 | 107 | 60.5% |
| rebelxt | 53 | 63.5% | 120 | 66 | 14 | 84.5% | 66.9% | 90 | 112 | 45.2% |
| malazan | 54 | 65.5% | 122 | 60 | 18 | 73.5% | 67.3% | 76 | 105 | 60.4% |
| fglider | 61 | 61.5% | 121 | 75 | 4 | 81.0% | 71.0% | 77 | 126 | 21.1% |
| eAshoka | 65-ish | 68.8% | 136 | 61 | 3 | 79.0% | 72.8% | 77 | 114 | 53.6% |
| Mistmanov | 77 | 69.0% | 133 | 57 | 10 | 76.0% | 69.7% | 97 | 109 | 66.7% |
| E323 | 78 | 60.2% | 115 | 74 | 11 | 71.5% | 76.9% | 78 | 105 | 18.4% |
| FlagCapper | 91 | 70.5% | 138 | 56 | 6 | 76.5% | 77.8% | 77 | 145 | 46.8% |
| overall | 57.7% | 2223 | 1608 | 169 | 80.2% | 61.0% | 84 | 104 | 44.4% |
I think that average game lengths for programs much stronger or much weaker than oddshrimp probably reflect relative strength. Those for programs near its own strength must reflect oddshrimp’s aggressiveness: It has the killer instinct, it wins fast. It’s striking how the win rate for long games varies so much between opponents. That probably says something interesting about style.
January 2014