Zezo forum: Just an idea for a new feature...

Just an idea for a new feature...

Post by jokap | 2021-01-17 | 15:45:26

Dear Cvetan, mastermind behind the router zezo!

Obviously it is not possible to find the ideal route just before a race starts as weather forecasts are changing every time they are published. Little differences can start to play out and the butterfly effect is taking over...

But shouldn't it be possible to calculate the ideal route from the past weather data retrospectively, i.e. once the race is over/the first boat has arrived, thereby creating something like an optimal route?

I think it would be a really intriguing feature for zezo to have we humans are really could at interpretation of visual data--as has been mentioned here and there in the zezo-cosmos. And after a race we are always trying to figure out what the winner did right andwhy he/she ended up in first place...
Machines on the other hand (e.g. your router) are extremely good at calculating optimal routes without being able to understand/interprete why they are the best routes.
So why not pair these two paradigms in order to learn something that potentially is yet to be learned, find routes to sail that only appear sensible in retrospective?

I think that would be a really fascinating feature!

To take it a tiny step further, one could also think about calculating this optimal route at every time step of the game, i.e. calculate a virtual leader that would be exactly where he is if he had have been able to anticipate all the weather updates...

What do you think, Cvetan?

Cheers,
jokap

6 Comments

Post by jokap | 2021-01-17 | 15:47:26

Please excuse all those typos :D

Post by zezo | 2021-01-17 | 16:16:18

I've done that in the VOR, for a long leg - China to NZ or whatever it was. The actual winner was quite close to the optimum route. Which is quite likely given 100000 boats with their own trajectories.

I don't think it would be very different in this race. Could save a hour or two overall, but with 500 boats finishing in the first hour and 1000 in the first two someone would find the better route if it existed.

P.S. Typos? What typos? I don't see much typos ;-)

Post by Ventura_317 | 2021-01-20 | 18:24:54

Dear Cvetan

I understand that during this experiment you simulated the travel of a boat with certain options following perfectly/exactly (HDG change time/ HDG /sail) prescripted by your own Zezo computed routing (updated every 6 hours and as soon as new data are available ) ..... and compared it with the Race Winner !

I beleive the gain on hours should not be so impressive as you mention it since you have already so many fans which are trying to do it on their computer ! Am I correctly understanding your comment ?

Thanks again for this so nice Router and your invaluable availability to reply quick and clear on the Forum to my more or less stupid questions !
Regards

Post by BooBill | 2021-01-17 | 17:10:15

A race with a million participants is in essence a monti carlo simulation. Almost certainly going to find the global optimum.

Which brings me to an enhancement that I've been noodling for some time. The limitation of the routing is that it is deterministic. It assumes the weather forecast will not change. Of course, it will, it's more of a stochastic process. Financial modelling of path-dependent derivatives has been tackling this problem for years. Monti carlo simulation is one method of estimating the intractable. There are others, but the key issue is going to be deriving volatilities and distributions of future weather.

Of course, if Cvetan can crack this nut, he'd immediately be poached by Goldman Sachs, lavished with millions of dollars and lost to us ;)

Post by YourMomSA | 2021-01-18 | 16:41:11

For this year's Vendee, the question being asked here really could just start with when the fleet arrived at Cape Horn. There was so much fleet compression south of Australia that nothing anyone did before that (short of making a mistake so enormous that you never caught up) was going to make a significant difference in the end result. If you look at the routes of some of the top 20-30 boats, you'll see some of them didn't do well Southbound in the Atlantic, and had significant deficits at Cape of Good Hope. Some also were casual about the short-gybing on the ice limit, accepting small losses there. None of that mattered because the weather brought everyone back up to the leaders. And that's what usually happens. Nothing surprising about it.

From there, I figure the best route was... West of Falklands... Then cross the fleet and head for that pressure ridge that killed Rostef. Then tack northbound earlier than the fleet and sail fairly close to Brazil. Continue Northbound, maintaining a position West of the main fleet. Then follow the router to the finish.

The winners did something close to that. I think the early tack North and routing near Brazil was the key move, as some of the boats who beat me were significantly behind me prior to that tack. I figure if you did it all perfectly, you might have come out an hour or more ahead of tigrou when the lead pack made the turn East toward the finish. But... From there on out, it was heavy fleet compression all the way to the finish. That advantage would have mostly disappeared, turning into a win by maybe 10-15 minutes.

Post by zezo | 2021-01-20 | 19:34:04

@Ventura_317

I just ran the router once with the 1 month of historical data that was used by the game during the leg. Which is not exactly simulating it because the router runs in larger steps and that may change the outcome a bit.