Show HN: Watch a neural net learn to play Snake

1 day ago (ppo.gradexp.xyz)

In browser PPO training demo, made possible by tinygrad: TinyJit -> WebGPU kernels.

Requires WebGPU.

34 comments

c1b

Reply
starshadowx2  1 hour ago

FYI this website sets off a bunch of Bitdefender alerts as being a suspicious web page. I assume probably false positives or something but still something you might want to look into.

"The page https://ppo.gradexp.xyz/ has been detected with suspicious activity. It is not recommended to continue browsing this website."

Same for:

https://ppo.gradexp.xyz/version.js

https://ppo.gradexp.xyz/dist/sizes.js

https://ppo.gradexp.xyz/dist/size_6/manifest.j

https://ppo.gradexp.xyz/dist/size_6/weights.safetens

https://ppo.gradexp.xyz/dist/sokol/demo.wa

ldoughty  10 hours ago

Really cool! But right as it was nearing 4,000, it seems to have corrupted itself and no longer got any scores above 0. Not sure if that's a code bug or a neural net issue.

avg500 -4.6 last 500 episodes

peak 3959.3 best window

roll/s 20.68 20-step avg

progress 4388 562749 episodes

  • c1b  8 hours ago

    Yes it just collapses eventually — never stabilizes. The training process is flawed, I suspect it has to do with the fact that some weights blow up over time, you can see in “weights” tab.

    But at around 4K avg score you should see it solve the env almost every time.

    Just a demo :) optimized for speed over stability.

    Reward structure: Step: -1 Dot: +100 Win: +1000 so ~4k is max theoretical score on 6x6.

    • ticulatedspline  5 hours ago

      maybe because it doesn't understand "done"? perfect play is impossible, random variance will cause scores to drop even if the model plays well and "wins". feels like it would get stuck in a loop trying to improve what can't be improved.

      2 replies →

  • r3trohack3r  9 hours ago

    I think I noticed it reach “end game.” The snake reaches a point where, if it gets any longer, it is out of squares and hits its own tail. So it finds the route through the squares that it can infinitely loop, never eats the ball, and score starts dropping and goes negative.

simedw  11 hours ago

Cool project!

I noticed that if you go from training to watch and then back, the training temporarily drop significantly in score.

  • bguberfain  9 hours ago

    It seems to be something related the moving average calculation. So it is just a glitch on the chart.

anthonycoslett  2 hours ago

Mesmerizing - could be its own digital art showcase XD Love what you've done here, friend. Looking forward to what you do next. <3

ziofill  6 hours ago

I noticed snake gets penalized for not getting to the apple early, is that what you really want? Snake is about how long it gets not about the balance between length and wall clock time

jesuo  6 hours ago

Poorly programmed, it doesn't learn from its mistakes, the games get stuck in a loop because the snake doesn't capture a piece but the piece remains and there's a gap, constantly moving the snake along the same path with negative scores in an infinite loop leaving an unaltered yin and yang ;) there's a repetitive pattern in these infinite games between the position of the gap and the piece

beardsciences  11 hours ago

My average eventually made it to about 3900, and then stagnated between 3600-3900. I'm curious if this is universal behavior or not. I'm up to about 5k steps.

neduma  11 hours ago

More details and implementation notes please?

  • cshimmin  7 hours ago

    It's on the page, if you click the little info icon in the upper-right. Here's the text but there's some nice graphics there too:

      Snake Game, training entirely in the browser. Built on tinygrad: the rollout / targets / train graphs are TinyJits authored in Python, then compiled once to WGSL and replayed here under WebGPU.

  Observation: flat 10×10 board (100) + 4-dim prev-action one-hot = 104 dims. fc_pi.weight is zero-init so the opening policy is uniform over the legal actions; fc_v uses tinygrad's default Kaiming init.

  Per rollout: T=24 × N=384 parallel snakes (9,216 transitions), then K=3 epochs × 4 mini-batches of PPO updates. GAE γ=0.99, λ=0.95; AdamW wd=0.01; ratio clip ε=0.1; grad-norm 0.5; Huber value β=1, val_coef=1; entropy bonus 0.008333333333333333.

  Action mask + value clip + KL early stop. The 4-dim prev_a obs tail lets fc_pi zero the U-turn logit (the env silently overrides same-axis reversals anyway). Value loss is max(huber(v_new−td), huber(v_clip−td)) at ε=0.2. Approx-KL is sampled after each epoch and breaks the loop at 1.5·kl_target.

jmclnx  10 hours ago

> WebGPU not available in this browser

Looks like this is for Linux and Windows, on NetBSD I get this issue :(

  • NoboruWataya  9 hours ago

    I got this in Firefox on Linux, just had to enable WebGPU in about:config (`dom.webgpu.enabled` = true).

    • jmclnx  8 hours ago

      Did not know that existed, I enabled it but no luck. Must be a NetBSD thing based upon this new message:

      > WebGPU is not yet available in Release or late Beta builds.

  • redshiftza  9 hours ago

    If you are using brave (which i assume also applies to chrome) , there is a menu at brave://flags , you can enable unsafe web GPU from there

austinthetaco  10 hours ago

my training on a 10x10 just randomly broke. i got to like 3600 then the graph went flat, the viewer on the left just showed it constantly restarting the game, and the scores in the negative. my average is now -10.