Neural Network Learns to Play Snake

this stupid group of circles are better than me at playing snake game FUCKKK!!

wouldn't it be simpler if it just did the logic to go in a straight line to x? this neural network is basically trying to brute force learn from a baby

I could count 25 neurons in input layer.. what extra information did the NN got?
Also, what does "Distance to it's tale" mean? The end of it's tale, the middle or any other part?

I wish you a speedy recovery and hope all god will makes everything goes well for you.

At the end it felt like the snake was synced up with the music and dancing along that was pretty groovy

What software is used to visualize the working of neural networks?

Left and right are mixed

8 directions? Like N NE E SE… ?

Trying to make my own with deep Q-Leaneing.

Still a bunch of mysterious parameters :D. We don't know why

TOP ! GRANDE

Channel of God: I trained some neurals that can build rockets 😂

Wow

Yea it could not learn the deep feature of not corralling itself with its own tail. Very interesting!!

Given your input layers it makes sense that it started to struggle when the worm got to big. It doesn't have the input layers to detect spatial availability like that.

When you see that each generation is better than the previous you know you actually made something
That's the best part of all the process of making this dumb ai

It cant get further because of lack on input! You give more on like how long the body is or where is the body now, and it will when

👍

So cool

I would not be adding more turns after eating food because this allows an inefficient strategy, instead I would fix the number of moves or remove more depending on how slowly the snake got to the food to reward more efficient behavior. It currently is doing enough which is good enough to survive for long and not die so it has no reason to change its behavior. It will only change if you reward certain things and this would be productivity efficiency. This is success per time so you'd limit the simulation and make the reward more varied for more complex and constrained behavior. That should let you train faster too.

Chat GPT is taking notes.

Super vidéo ! Parfait pour un pic nic entre potes pour se relaxer. MERCI

Hello sir, how can you visualize that progress in real-time.

how would you code this?

great video thank you

How about trying snake with no walls?

Failed. downvote.

just take a snake game and put it one player mode and you can challenge the AI

Can you try starting to train the model with larger length similar to the length at the last death?

Woooooo 😮

1

Still better than my efforts!

How do you display that neural network from the side?? I want to make A.I.

Is this how neural networks look in real-time. Every time I look for how data is fed into machine learning and does things but I don't get a single video. It's all just a theoretical part

Which activation function was used ?

There is one extra output neuron in the model, the snake can only ever move in 3 directions straight, left and right.

It really loves that top left corner

Hey Man! Loved this, I want to connect with you for some amazing gig, I don't know how to do this, can you please drop an email or something where I can dm you!?

I think the next strategy it has to learn is making space to go and that requires some serious training with very specific rewarding system

Maybe reinforcement learning with lower amount of movements will yield better results?

Are dots appearing randomly? or its a match fixing?

hi, i am trying to adapt a model that does the same job on a different game i wrote on python but i couldnt find a proper guide that explains it from scratch, if u could link any support please do

How so win if still many glitch and line error? 🥱🥱

at some point it just became a loading screen

i wonder if the snake would have evolved to the point of getting so overweight it hires someone to bring the food to him cause he can no longer move

funne that until generation 17 the snake only try to run clock wise, since it was the frist attempt that led to positive results.

Why not try recorrent network? Is what is has been used on self drive cars.