lsusr comments on The Darwin Game

[lsusr]

I will not disqualify anyone who can complete 100 move calls in 5 seconds (0.05 seconds per call on average) on my T450s.

I will not disqualify anyone who can complete 100 move calls in 0.05 seconds on my T450s.

Edit: Corrected time limit downward by 100×.

[philh]

Thanks. I confess I’d been hoping for more like 100x that, but not really expecting it :p

[lsusr]

My previous answer was a mistake. It is actually 100 move calls in 0.05 seconds. Sorry.

The numbers are brutal.

If a game goes for 50 rounds and 10 different bots each use 5^[1] seconds per move and there are 550 moves per bot per pairing then it would take 4.36 years to run this tournament.

$$\frac{5\frac{\text{seconds}}{\text{move}}\times 550\frac{\text{moves}}{\text{spawning}}\times 100\frac{\text{spawnings}}{\text{bot}\times \text{round}}\times 50\frac{\text{rounds}}{\text{tournament}}\times 10\text{bots}\times 1\text{tournament}}{31,536,000\frac{\text{seconds}}{\text{year}}}=4.36\text{years}$$

However, do note that the guarantee is “0.05 seconds per 100 moves” as opposed to “0.0005 seconds per move”. If you only have to run expensive computations once then, depending on your algorithm, the right caching system might get you close to 100× performance.

I can add caching functions to the extra package if that would help. The computer in question has 4 processors and runs Linux so multiprocessing can get you up to a 4× speedup.

---

1. ↩︎

   500 seconds per move is 100× my original limit which equals 10,000× the corrected limit.

[philh]

Oh, geez. I figured it would be too long, but I didn’t think about just how much too long. Yeah, with these constraints, even 5s per hundred moves I agree is unreasonable.

Caching seems easy enough to implement independently, I think. No need for you to add it.