alpha-zero-boosted VS katrain

> I came up with a nifty implementation in Python that outperforms the naive impl by 30x, allowing a pure python MCTS/NN interop implementation. See https://www.moderndescartes.com/essays/deep_dive_mcts/

Great post!

Chasing pointers in the MCTS tree is definitely a slow approach. Although typically there are < 900 "considerations" per move for alphazero. I've found getting value/policy predictions from a neural network (or GBDT[1]) for the node expansions during those considerations is at least an order of magnitude slower than the MCTS tree-hopping logic.

[1] https://github.com/cgreer/alpha-zero-boosted

What you can do is checkout the algorithm at a particular stages of development. AlphaZero&Friends start out not being very good at the game, then over time they learn and become super human. You typically checkpoint the weights for the model at various stages. So early on, the algo would be like a 600 elo player for chess and then eventually get to superhuman elo levels. So if you wanted to train you can gradually play against versions of the algo until you can beat them by loading up the weights at various difficulty stages.

I implemented AlphaZero (but not Mu yet) using GBDTs instead of NNs here if you're curious about how it would work: https://github.com/cgreer/alpha-zero-boosted. Instead of saving the "weights" for a GBDT, you save the splitpoints for the value/policy models, but the concept is the same.

I like Sabaki, but there's also GoWrite, CGoban (the KGS client), and others (search for "SGF editor"). You can also review .sgf with OGS online, or with AI Sensei. KaTrain is a very good AI client that can review .sgf as well.

The easiest way to install & use KataGo is via KaTrain: https://github.com/sanderland/katrain

I bring in my laptop with KaTrain installed and ask if my opponent is down for entering in the moves as we play as reviewing after. Its fun to check out different models, analysis depths, and local areas

I run KaTrain on my older Intel Mac with no issues. The OS is older, 12.6.3 (Monterey). I use the KaTrainOSX.dmg released by the KaTrain project. How did you install KaTrain?

If you have a good computer you can run katago locally using something like katrain which makes installing it simpler and comes with options like controlling katago's strength.

If you are set on using Sabaki specifically I can't offer much advice since I don't have much experience with it, but what I can say is that I personally use Katrain to run Katago and I highly recommend it. If you want to try it out, you can just go here for the latest version as of today. Just scroll to the bottom and download "katrain.exe". It has many many very useful features - and of course will always evaluate the move that was actually played.

You can create an issue in https://github.com/sanderland/katrain/issues. The author actively resolves issues there.

I use KaTrain. It's great. https://github.com/sanderland/katrain/releases