Since my faithful old notebook decided to pass away I bought a new one and so I am able to play with Lizzie and LZ. Since Lizzie doesn't have a comfortbable menu or explantions I wanna ask some questions:

1. I have Lizzie 0.6. ... does it update automatically or do I have to look for new updates?
2. Does Lizzie 0.6 look for the newest LZ autoamtically or do I have to look out?
3. What do the colors of the move proposals mean?
4. What does the number in the move proposals mean? I think the higher the number the higher/deeper the moves evaluated by LZ, but what reference e.g. does a number like "100" have...like 100.000 moves watched ahead by LZ?

1) It doesn't auto-update
2) It doesn't by default, and I'm not sure there's a setting for this.
3) Blue = LZ choice at this point, Green = other moves really considered, -> red = moves much less considered
4) The top number is the estimated winrate of the move, and the lower number is the visit number of the move. Basically, it's the number of time a sequence starting by this move has been evaluated. At each visit, it evaluate a new move deep down in the tree, growing step by step the explored tree.

I've downloaded Lizzie0.6windowsx64GPU. As far as I understand it is Lizzie 0.6 with LZ 0.16, network #157. Am I thinking correct if I want to get #214 to basically download it and rename it to "network.gz" or is it more complicated to update?

You're perfectly correct .

I made some videos about using Lizzie, here's the first version: forum/viewtopic.php?f=10&t=16475 with Lizzie UI explanation starting at 12 minutes in the first video.

Thx a lot, very good stuff, esp. for greenhorns.

I have a question about Lizzie's UI. On the win percentage graph, there is also a red line that fluctuates to *some* degree, just like the green line that reflects the actual game. What is the red line? I've never understood what it shows.

Do you mean the red bars which go up or down from the middle line? These are "blunder bars", and show how big the mistakes were of the mainline move versus LZ's best move. So a bar down means black made a mistake and the winrate green line goes down, bar up means white made mistake.

Here's another interesting one if you wanna play LZ vs LZ. You can do it by just always pressing enter and LZ will play a move. Does it matter how fast I press enter or not? Also is there any way to let LZ play itself automatically?

I'm still using two gtp script from gogui, but there are better solutions

you can try https://pypi.org/project/gomill/#description even though I haven't tried it myself

A question about the diagram on the left side. What do the one line below/above the x-axis mean? Why is it there? Does it mean that within that boundary the game is still open, outside basically lost?

All bots are fallible, otherwise the only winrates would be 100% and 0%.

In this case, note the small number of visits in the red circle. That shows that LZ tried the move, realized it was a disaster, and stopped trying it. The 80% winrate estimate is what it had calculated before abandoning the play. Humans would probably change the winrate win the error was discovered, but bots do not base their decision whether to play a move or not on win rates, so they do not bother.

Since the display is for human consumption, they might as well not show the move at all, but for the fact that sometimes bots dismiss a move too early, and a human might decide that such a move deserves more investigation. That's not the case here, OC.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

oh, okay. The red bars were what I was referring to. Thank you! I was always wondering what it was showing me.

What is its expected line after, does it see the capture?

Note that Leela Zero didn't actually suggest the move should be played; it liked another move 30 times as much. It just looked at it. As Bill says, once it realizes it couldn't possibly be the best move, it doesn't have any incentive to go back and calculate exactly how bad it is.

Ok, that makes sense, but isn't it weird for us users? If I see a move suggestion with 80+ win probability, even in red, i.e. not prefered by LZ, I think it must be still very well be playable. There are plenty of examples where such moves are indeed playable. But it's good to keep that in mind: if it's 'red' it's 'at your own risk'.

As the third most visited move it was suggested by the policy net at least. I doubt an ELF net would consider it for example. But LZ has a specific defect in its training wrt some ataris and move legality, which may be in work here.

To get a more accurate assessment of a move, play it and let LZ think for a while (e.g. until 4k visits like the #1 move had before). You will likely see that 80% drop pretty fast. But yes that LZ can have super-human instinct at 10 playouts in some positions but atari blindness a 25k won't (often) do at >100 in others is the wonders of neural networks . See viewtopic.php?p=242513#p242513 for another example of atari blindness.

For better or worse, Leela Zero is designed to play as well as possible, not to teach people how to think about positions. Certainly one could argue, and I would agree, that it has accomplished the first goal well enough that it makes sense to transfer a lot of our energy (as programmers and researchers) to the second.

You have to remember that the win rate estimates of moves with low playouts can be way, way off. The bots do not use win rates like humans want to do.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins