Post by Disposable1234 in Idle Machine Learning comments

Viewing post in Idle Machine Learning comments

It is so terribly annoying when they just spend 50 generations bashing their heads against the exact same rock.

You have plenty of steering, SO STEER. Stop speeding at the rock, YOU HAVE A BRAKE SO USE IT.

Why are you actively steering INTO THE WRONG DIRECTION after 30 generations of suicide? I lowered your top speed so you'd have a better chance of NOT HITTING THE EXACT SAME ROCK AS YOUR GREAT*30 GRANDFATHER.

I know the game CLAIMS they learn better after retiring once and getting a 2nd sensor, but these lemmings only seem to get dumber. Maybe that 2nd sensor is more curse than blessing?

vashgard1 year ago(+1)

hahahaha I made this game many years ago, and I still enjoy getting comments on it! ^^

Maybe it was badly explained on the game (I don't remember now) but second and third sensor don't make it learn faster, otherwise it makes they learn slower, but able to learn more and better (in machine learning terms, they have bigger layers).

Sometimes learning process gets stuck on a "local minimum" and it is very hard to get out of that. The only advise I can give you is to made they to "unlearn" a lot to get away of that "local minimum", and then start again the learning process. To do that, get back to old maps for some generations, or change the configurations by A LOT.

The green bike is there to allow you to see if each generation is evolving or they are stuck on a local minimum and getting away from it: if the green bike is always doing the same, they are not learning, neither "unlearning".

Another hard option you have is to retire early, you won't get the bonuses from completing all races, but you will start again with the same amount of sensor to try again learning from 0...

haxihoovis136 days ago

Isn't the term local maximum? It's the max you can get, but if you go back a few steps you can do better

vashgard136 days ago

Good question!

You are right, in this context would be a "local maximum", because they are stuck on "lap completion %", something you want to maximize but you can get stuck on a "local maximum".

I used "local minimum" because I was thinking on "lap completion time", something you want to minimize but you can get stuck on a "local minimum".

In short: You can use Maximum or Minimum depending of what are you measuring.

itch.io

Viewing post in Idle Machine Learning comments