BetaTetris - 219 Lines on 29 Start (12 Hz tapping, no adjustments)
Рет қаралды 495
2 жыл бұрынThis is an slightly improved version of BetaTetris (a machine-learning based agent of Tetris I made last year), and it is pretty good at level 29 playing now!
The average performance of this version is about 43k with 12 Hz & no adjustments. I ran this version for about 300 games and it got over 210 lines in 2 of the games, so it is roughly an 1-in-150 luck.
@CheezTetris 2 жыл бұрын
This is very interesting! I've been wanting someone to make an AI like this for a long time, especially since stacking is an aspect of killsceen that everyone seems to ignore right now. I would love to see more of its playing in the future
@BunnyBoing Жыл бұрын
fun to see that slow tappers (or tappers in general) can play on killscreen.
@codenamelambda 2 жыл бұрын
It honestly seems to me like the machine learning approach yields roughly the same results as just learning at as a human would [minus input accuracy, that's not as much of a concern for the AI], though I'd expect the "search for strategies" to be much less directed, in a way, where people would think forward to possible situations and learn how to approach a similar situation without thinking through sheer experience after enough games, the AI just stumbles into these approaches I'd guess? Or does this AI maybe subtly manipulate RNG as well?
@adrien_1018 2 жыл бұрын
Yeah, machine learning is more like inducting the strategy from millions of games it played. This agent cannot manipulate RNG, as I trained it on a different and more random RNG that approximately matches the original piece distribution.
@jame2358 Жыл бұрын
PЯӨMӨƧM
Bon Bon Media
Рет қаралды 12 МЛН
Adrien Wu
Рет қаралды 16 М.
Max Buck
Рет қаралды 221 М.
ВЛАДУС ИГРАЕТ
Рет қаралды 907 М.