Jashank Jeremy's Blog

I've been a bit remiss in keeping this data up to date. Here's three fresh new data points as we roll into round 21.

round 17 ran at 20:01 on 31 Jan. 43 submissions, 42 draculas compiled, 42 hunters compiled, 1722 games ran; fastest game 750 μsec; lower quartile 310.45 msec; median 1.11 sec; mean 3.10 sec; upper quartile 4.02 sec; slowest game 58.28 sec.

round 18 ran at 06:19 on 1 Feb. 44 submissions, 43 draculas compiled, 43 hunters compiled, 1806 games ran; fastest game 756 μsec; lower quartile 294.27 msec; median 1.12 sec; mean 3.10 sec; upper quartile 4.01 sec; slowest game 48.25 sec.

round 19 ran at 20:01 on 1 Feb. 48 submissions, 48 draculas compiled, 47 hunters compiled, 2188 games ran; fastest game 1 msec; lower quartile 303.18 msec; median 1.09 sec; mean 2.96 sec; upper quartile 3.70 sec; slowest game 56.32 sec.

round 20 ran at 06:19. 49 submissions, 49 draculas compiled, 49 hunters compiled, 2330 games ran; fastest game 188 μsec; lower quartile 303.12 msec; median 1.08 sec; mean 3.18 sec; upper quartile 4.06 sec; slowest game 69.41 sec. in exactly one game, all AIs broke in the first round..

That band in the middle is coming back tighter again, as the width of that middle 50% range is shrinking again; we've come off from the peak in round16 of a range of +3.95s in that middle 50% down 15% to +3.40s in round19, but that's surged again.

Next round at 19:59. Roll on, non-daylight-savings time.

Round at 06:18. 43 submissions, 43 draculas compiled, 43 hunters compiled, 1806 games ran. The count of hunters and draculas is up again, pushing us up against a half hour to run all the games.

Fastest game 1.22 msec (!!!); lower quartile 285.93 msec; median 1.10 sec; upper quartile 4.24 sec; mean 3.12 sec; slowest game 44.02 sec:

And given it's been a while since we saw the shape of progress, let's see what the overall trends have been. Here, again, is the range graphs showing slowest, quartiles, median, and fastest games. The inner section is the most interesting, especially its slow trend upwards. I suspect the slowest/fastest games tracking is disappearing into noise, given its variability.

And here's just the inner section, with the mean line. The mean has been well and truly below the upper quartile, suggesting the average games is relatively faster, though the size of the range is still slowly increasing.

Here's today's weird graph: I've got the candlesticks, showing that range and the movement round-on-round of the median, and have overlaid, on the right (linear) axis the size of that 50% range. The resulting graph is a hot mess, and Edward Tufte would probably kill me for it, but:

It's not just my imagination; that middle 50% has been creeping upwards at a scary pace.

Next round at 20:00 sharp.

40 submissions, 40 draculas compiled, 40 hunters compiled, 1560 games ran. We're back to 100% compiling, and the number of draculas and hunters is up again; hooray!

Fastest game 580 μsec; lower quartile 240 msec; median 971 msec; mean 3.04 sec; upper quartile 3.75 sec; slowest game 69.24 sec.

Gal also requested a distribution, which I'm happy to distribute. I've tweaked 10% of values as outliers, so each bucket is 0.08s wide.

Next round 20:02.

Round at 06:14. 36 submissions, 34 draculas compiled, 35 hunters compiled, 1156 games ran.

Fastest game 766 μsec; lower quartile 270 msec; median 992 msec; mean 2.75 sec; upper quartile 3.48 sec; slowest game 79.16 sec. As expected, the timing curve is basically as it was last night, though it's crept upwards a bit throughout, suggesting a general slowdown, and not a player slowdown.

Next round tonight at 20:01.

Round at 20:01. 36 submissions, 35 draculas compiled, 35 hunters compiled, 1190 games ran. Record!

Fastest game 739 μsec; lower quartile 248 msec; median 974 msec; mean 2.60 sec; upper quartile 3.21 sec; slowest game 65.50 sec. The graph:

(For those of you whinging about the rendering difference from the usual graphs: this is what Google Sheets looks like on my workstation, in Firefox 60.0a1.)

The middle-50% difference isn't quite as obvious just looking at just the time graphs; however, here's some pretty candlesticks to illustrate what's actually been happening over the past few rounds, just showing that middle 50% range in the lo..hi:

My conclusion? I don't think this is just slow hunter/dracula strategies any more; some of the (Game|Drac|Hunter)Views are apparently terribly slow. I want to do some profiling and benchmarking of groups' current views to see what they do, only in the process of creating and destroying the view.

Also: here's two interesting graphs that I've been promising for a while, showing locations on the map in round 13 ordered by move popularity on a log scale. Galatz and Klausenberg are, by a long shot, the most popular places to be if you're a Dracula, and Castle Dracula and Klausenberg if you're a Hunter.

This doesn't count Dracula's HI/Dn/TP moves, because that'd require I write another trail parser, and I'm honestly a bit sick of doing that at the moment. I've even quartered the hunter move count... those are actually that far beyond the rest of the graph, even on a log scale.

I'm tempted to do this over all the rounds thus far, to see what the shift looks like over time. My suspicion? Very little, given the fastest games are still those played in CD, GA, KL, BD, and SZ, and I suspect it's only that side of CD because hunters don't like coming in from the Black Sea

Next round 06:14.

Round at 06:13. 34 submissions, 31 draculas compiled, 32 hunters compiled, 961 games ran; the game engine broke 36 games by silently failing before the first player could move. Those games have been excluded from stats and from the rounds, and in particular this has affected groups g014890, g014891, and g014895.

Fastest game 3.94 msec; lower quartile 253 msec; median 1.15 sec; mean 2.76 sec; upper quartile 3.78 sec; slowest game 54.06 sec. Games are still slowing down, and the spread across the middle indicates some quite complex 'thinking' happening inside AIs.

I'll update here once I find out what's happened to those 36 games.
Update: I have no idea what happened to those 36 games. I can't reproduce any of the failures I saw. I've disabled five hosts from the runner cluster which caused problems; I think we're good now.

Next round is at 20:01.

Round at 20:02. 34 submissions, 32 draculas compiled, 31 hunters compiled, 961 games ran. Fastest game 5 msec; lower quartile 222 msec; median 901 msec; mean 2.45 sec; upper quartile 2.84 sec; slowest game 41.25 sec. It's sped up a bit, and the curve is flatter again thanks to more consistent games.

Still a better competition than the Hottest 100. Next round 06:13.

Round at 06:13. 32 submissions, 30 draculas compiled, 30 hunters compiled, 870 games ran.

Fastest game 4.15 msec; lower quartile 221.77 msec; median 852.32 msec; mean 2.53 sec; upper quartile 2.78 sec; slowest game 70.50 sec; yesterday's performance spike wasn't just an odd-ball, as it's tracked into today as well.

It looks like there hasn't been much change overnight, so the change looks pretty flat, overall, but there's definitely something happening, especially with the slower games; there's been a spike in the range of the middle 50% of games, even though the overall median game time is coming off; the maximum game time is also up somewhat, though not significantly.

Next round tonight at 20:02.

Round at 20:02. 32 submissions, 30 draculas compiled, 30 hunters compiled, 870 games ran.

Fastest game 3 msec; lower quartile 224 msec; median 897 msec; mean 2.31 sec; upper quartile 2.50 sec; slowest game 54 sec. It's good to at least see the mean is back in the middle 50% of games.

Not as obvious just by the numbers, but definitely visible on this graph and the other graphs I've posted recently: games have gotten noticeably faster across the majority of the curve. I have absolutely no idea what's caused that sudden jump in performance.

Next round at 06:13.

Round at 06:12. 30 submissions, 29 hunters compiled, 28 draculas compiled; 784 games ran. Most of yesterday evening's compile errors have gone away, thank goodness, so we're back on the upward trend for games.

Fastest game 748 μsec; lower quartile 305.42 msec; median 1.11 sec; upper quartile 2.76 sec; mean 2.89 sec; slowest game 66.82 sec.

Games are still trending slower, which I fully expect, and the mean is still diverging away from the middle 50% of game times, which is very peculiar, and I'm not sure quite what to make of that. Also of note is the jump in the lower quartile time, suggesting that more strategies are appearing.

I'm tempted to do some sort of move analysis -- what are the common locations on a map, where do people encounter traps or vampires, etc. -- but I fear that may influence strategy a bit too much. Decisions, decisions.

Next round is at 20:02. Stay cool.