story
Yes, selecting 1000 and then cherry-picking would be a great idea in theory, but ghost data is quite heavy (around 300–400 KB per ghost). We are already compressing it as much as possible: saving only about 10 frames per second and interpolating the rest, reducing decimals to a minimum, and storing rotation only on the Y axis. We also keep only the ghosts of cars that complete a race in under 4 minutes to avoid excessively large files.
Because of this, loading 1000 ghosts and then picking a suitable set of 7 close to your skill level would make the process too slow. Currently, we have a maximum load time of 10 seconds for the ghosts, and we are aiming to ensure the game works smoothly even on slow connections.
For example instead of storing, x1, x2, x3, x4, x5, you can store x1, x2-x1, x3-x2, x4-x3, x5-x4 Using few decimals it may compress better with the standard algorithms.
Assuming almost constant speed, you may guess x3 from x2 and x1 and x3=~ 2*x2-x1, so instead of x3, x4, ... you store x3-2\x2+x1, x4-2\x3+x2, and compress it.
For a car, I'd assume constant acceleration and if I didn't make a mistake, x4=~ 3x3-3x2+x1, so you store x4-3x3+x2-x1 and compress the sequence.
In any case, the ideal solution would be if the API allowed us to send the player’s average time as a parameter, so we could fetch opponents at a similar skill level, that would greatly improve the gameplay experience. I’ll mention it again to the backend team, and if we’re lucky, maybe they can prioritize that feature.
Cheers, and thanks again!
x4-3*x3+x2-x1 --> x4-3*x3+3*x2-x1
Good luck with the project.
PS: Usually tweaking the parameters to select the opponents/ghost and make the game fun (that is what is important) is hard. (More cars? Less cars? Use the last time of the player? Use the average time of the player?) So after you solve the problem, a post-mortem may be an interesting blog post to summit here.
PS2: another typo in my first comment: defectively -> effectively
