Hacker News Reader

Comment by JoshTriplett | Hacker News Reader

At generation 83, a single bird emerged that successfully navigated through several hundred columns (current score 100000+) and showed no signs of failing.

It took a few dozen generations to find versions that would make it through a few columns if those columns had gaps without too much vertical distance between them. Somewhere in generation 60-70, I could see versions figuring out how to transition between heights, but failing by overshooting, or not accounting for gravity on the far side of the gap. But once it figured out how to transition between distant heights, it had something that kept working.

GIF of the successful bird: https://imgur.com/a/A0li8

View on HN

53 for me. The previous generations didn't even look close then all of a sudden this one.

Same. Mine is on generation 22 and is the first one to get past even 30 seconds. So far no sign of stopping.

edit: Gen 22 still going strong @ 1 bird, up to 231,000 points. Looking at game.js it's a fairly simple model, I suspect once you get a bird that makes it past the 1-minute mark it will be able to continue forever.

2nd generation for me. Very suspicious.

EDIT - to be more clear: the first generation died in a couple of seconds. A bird from the second generation was immediately able to navigate forever. Either this was my chance to win the national lottery (what a way to waste it) or something's wrong.

For me it was 1st generation and I was wondering what this project was all about. Until I hit F5, now it's 50 generations and still to successfull bird in sight.

What's interesting is I've run it several times now, and I've gotten the perfect bird about half the time on an early generation (<10), but sometimes it just doesn't want to find it for a while. I'm wondering if the net is just at the threshold of being the minimal set of neurons that are needed for it to be able to get it so easily by random chance.

I was surprised to get a winner on generation 7 the first time I started the trainer. Subsequent trials were much worse.

I was surprised that I got one that get really far and then in the next generation they failed right at the start

In a GA approach, you'd have kept that bird AND mutate/breed offspring from it. In a NN, you often get oscillation during training.

This is a GA. I just think it doesn't practice elitism (saving the best solutions found so far.) Or it's nondeterministic.

The problem seems to be the map.

It's always different and you can get to 1000 points if you are lucky with the map. But the one that got to 1000 might not get a gap that didn't show up in the last map.

Wow, I guess I was very lucky.

I am running right now a bird from generation 3 going at 26k+ score.

How is it possible to get such good specimens after so little generations? I'm guessing part of it is the simplicity of the model and the environment (the game), but still.

gen 6 seems to have gotten it right for me, still running, ill follow back when it loses edit: still running at 4600+ 5 minutes later on gen 6

took 230 generations for me to do anything then it just went from dying within the first 3 pipes to mastering it within 4 generations

On my first run, 2 individuals learnt(?) perfect play in 3 generations.

On my second run, 3 individuals seemed to have learned perfect play by gen 8, but after successfully navigating through dozens of pipes, two of them died and one continued indefinitely.

Similar experience here. Went from dying within the first 3 pipes to a perfect run at generation 68.

Gen 17 had its last bird run forever

Generation 10 for me

gen 9, lots of close calls, but still flying fine

57 for me.

Reply on news.ycombinator.com