#mon2y — Public Fediverse posts

`python monty.py c4 train -e 1 -i 2000 -e 10 1053.11s user 54.03s system 94% cpu 19:37.66 total`

`target/debug/mon2y_rs c4 --iterations 2000 -p m,m -e 10 103.27s user 5.87s system 92% cpu 1:57.50 total`

Single thread only so far, and added some cacheing, but _yikes_ that's an improvement in speed.

#rust #mon2y

My code is somehow teasing me.

See that UCB at the very end there? The one that's negative? That's the only good next move.

I bet it's a tiny thing somewhere too.

#mcts #mon2y #programming

My #Rust MCTS implementation is actually returning actions! Haven't measured yet, but it's looking like it's speed is very very very good compared to the previous Python implementation.

Unfortunately, it also seems to be returning also exactly the worst possible move each time. It's like that scene in Enter The Spiderverse: You can't get 0% on a multiple choice quiz without knowing the correct answers.

Now, I've just got to figure out where it's flipping.

#mon2y

Almost got enough of the MCTS implemented in the #rust rewrite of #mon2y to hook it up to play connect 4. And maybe then start using it for something useful (ie - help me playtest designs). Or, spend the time of getting it multithreaded).

It's been a good learning experience, but my attempts to make it very generic have made it a tough first Rust project for somebody who hasn't touched a language lower level than Python in 4-5 years.