AI Thinking Timer Now Live! 🧠⏱️
Go Game Online with KataNet AI » Devlog
Hey Go Strategists! 👋
Just rolled out a neat little update to make your AI battles (and hint requests!) a bit more transparent.
What's New?
- See the AI Think! Ever wondered just how long that clever AI is mulling over its next move? Wonder no more! Now, when the "AI Thinking..." panel pops up (or when you ask for a hint!), you'll see a handy timer ticking away. This gives you a better idea of the processing time, especially if you've set the AI to be a real brainiac in the settings! 🧐
This was a requested feature to give you a bit more insight into what's happening behind the scenes during the AI's turn.
As always, your feedback is super valuable. Let me know what you think, and happy stone-placing!
Files
go.zip Play in browser
45 days ago
Go Game Online with KataNet AI
A versatile Go board: Play, analyze, and challenge AI opponents.
More posts
- Your AI Go Coach Just Got Smarter!37 days ago
- ✨ Learn Go Faster with Your New AI Coach! ✨40 days ago
- Know Your Odds! AI Winrate Now Powers Up Your Go Analysis!41 days ago
- Accessibility Upgrades! ♿41 days ago
- Supercharged Review Mode! 🚀42 days ago
- Level Up Your Game Review! New Analysis Tools Added!42 days ago
- AI vs AI Gets More Interesting & Safer Plays!42 days ago
- Auto-Save, AvA Pause, SGF Power-Ups & More!42 days ago
- Take Control! "Play From This Position" Gets a Major Upgrade!43 days ago
Comments
Log in with itch.io to leave a comment.
The new timer is very useful. It confirms what I suspected: that the AI thinks for much longer than the limit set in preferences. This time can now be very long, even in the early stages of a game played on 9x9.
After that update I uploaded another, fixing an error that MCTS Iterations (Total) even set below 10000 iters was not working, now is fixed.
So when you set it to 5000 actually it had been doing 10000 iters.
Try very low iter say 500, it finishes very fast. But the results are not what you would expect.
The limit is NOT total time it is Time Limit per Evaluation Batch (ms) i.e per worker
Nah, I was setting the iteration limit to 5000 like you said. Now I've also set the time limit lower. It plays quickly enough to be usable again. Too bad it's also weak like before.
There was a problem with the time limit however, I have just uploaded the fix. Now I think everything should be OK.
Yes, less iterations more Dummy ;)
Now you can play purely "on time" set MCTS Iterations (Total) = 1 000 000
Time Limit per Evaluation Batch = 15 000
your move should be 15s (little more) (you will see how many iters your computer does for 15 seconds)
Now it consistently thinks for about twice the time limit, which is perfectly fine, but I can't set the iteration limit that high because it starts freezing my browser again. And even at 150K iterations it doesn't seem any smarter than at 5K.
You may set 1 000 000 iterations in options but what these numbers show is what actually was calculated, in this case:
9951 of 10000 moves. If you set low time you will never get to 1 000 000, it will stop on the time.
You may look at the Help there are tips there.