Leela Zero vs Leela
-
InsideLiberty
- Beginner
- Posts: 6
- Joined: Tue Jan 15, 2019 5:42 am
- GD Posts: 0
- Has thanked: 5 times
- Been thanked: 1 time
Leela Zero vs Leela
Sorry if this question has already been resolved, but is there a consensus on whether Leela or Leela Zero is better for SDK game reviews? I've seen some back and forth online arguing that even though Leela is weaker, because it is human trained, its better for learning if you are below dan level. Thanks!
-
Uberdude
- Judan
- Posts: 6727
- Joined: Thu Nov 24, 2011 11:35 am
- Rank: UK 4 dan
- GD Posts: 0
- KGS: Uberdude 4d
- OGS: Uberdude 7d
- Location: Cambridge, UK
- Has thanked: 436 times
- Been thanked: 3718 times
Re: Leela Zero vs Leela
Not sure who decides if there is a consensus! I'd say give both a go and see which one gives you advice you can understand. I think Lizzie is the best tool for interactive reviews and that's for LeelaZero.
-
Bill Spight
- Honinbo
- Posts: 10905
- Joined: Wed Apr 21, 2010 1:24 pm
- Has thanked: 3651 times
- Been thanked: 3373 times
Re: Leela Zero vs Leela
In practical terms, there may not be much difference between the two for SDKs reviews. In either case I would suggest looking at, say, what the bot thinks are the three biggest mistakes and trying to understand why. Using the program to play through variations can help understanding, not just observing any lines that the program used to make its decision. Since the program is so strong, the player will probably need to explore the game more to see why the program made its assessment.
That said, I would recommend Leela Zero or Elf or other top bot. Leela 11 makes demonstrable mistakes. Also, IMHO SDKs have started to form habits and opinions, so that they see some good plays but overlook others. The bots that have been trained on self play are more likely to suggest good plays that would never have occurred to SDKs, and that is a good thing at that stage. SDKs are starting to build their ability to calculate variations, but what good is deep reading if you don't see the best possible plays and replies?
Edit: Corrected typo. The last word was replays, but I meant replies.
That said, I would recommend Leela Zero or Elf or other top bot. Leela 11 makes demonstrable mistakes. Also, IMHO SDKs have started to form habits and opinions, so that they see some good plays but overlook others. The bots that have been trained on self play are more likely to suggest good plays that would never have occurred to SDKs, and that is a good thing at that stage. SDKs are starting to build their ability to calculate variations, but what good is deep reading if you don't see the best possible plays and replies?
Edit: Corrected typo. The last word was replays, but I meant replies.
Last edited by Bill Spight on Sat Feb 09, 2019 9:39 am, edited 1 time in total.
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins
Visualize whirled peas.
Everything with love. Stay safe.
At some point, doesn't thinking have to go on?
— Winona Adkins
Visualize whirled peas.
Everything with love. Stay safe.
-
InsideLiberty
- Beginner
- Posts: 6
- Joined: Tue Jan 15, 2019 5:42 am
- GD Posts: 0
- Has thanked: 5 times
- Been thanked: 1 time