Hugging Face's logo

Join the conversation

Join the community of Machine Learners and AI enthusiasts.

Sign Up
Back to feed
Chris4K 
posted an update 5 days ago
Post
3,016
Can a network whose trunk is only XOR, table lookups and integer adds learn Connect Four?

Partly. I trained a GF(2⁸) integer model (8.4M params, CPU-only) on 320k games labelled by a negamax search teacher.

What it learned:
• agrees with the teacher on 59.6% of training / 41.0% of board-disjoint holdout positions
• beats a random player 85.2% of the time

What it didn't:
• loses to a 1-ply "win now / block now" rule (wins 37.2%)
• loses to 2-ply lookahead (wins 24.3%)

Agreeing with a teacher is not the same as playing well, and this release measures both.

Caveat up front: the float32 read-out is 99.6% of the parameters. The trunk is float-free; the full model isn't yet.

The repo also documents six bugs that each produced a plausible but wrong number. In one, 41% of the supervision was silently all-zero.

🎮 Play: Chris4K/connect-four-gf256 (Space)
🧠 Model: Chris4K/connect-four-gf256
📦 Dataset: Chris4K/connect-four-gf256 320k games + per-ply targets, MIT
💻 Code + paper: https://github.com/Ckal/connect-four-gf256

Feedback welcome, especially: which baseline would convince you?
dipankarsarkar

The 1-ply rule is already the convincing baseline. It also tells you what to measure next.

"Win now / block now" only decides when a threat is on the board. In quiet positions it has nothing to say.
So wrap your net in that same override and play it against plain 1-ply.
If it wins, the trunk learned positional play. If it only ties, the 41.0% holdout agreement is coming from tactical positions the rule already covers.

You can see the same thing without playing a game: split the holdout agreement into threat-on-board vs quiet positions.

Which half do you expect to be higher?
Lounafa1

Wow it's interesting! I love experimenting with unusual neural network architectures as well
In this post