Open pondturtle opened 5 years ago
A(G)Z and LZ's neural net takes the last 7 moves as the input, so if you don't feed it with a reasonable history it may not come up with reasonable analysis, so this is not so trivial to implement (same for Leela Chess).
@alreadydone i'm not so sure about that; for as Henry Ford said adroitly, "History is Bunk" and, if anything, having a history would distract LZ from acute observation.
@pondturtle As to wanting editor features, Sabaki already has them;
and i haven't yet tried it out, but FeatureCat has a Sabaki fork in Lizzie style called Shell and Slate. In fact, that might be such a good idea, i'm gonna check it out right now :) PS Oh - it's only a change of look, not a change of feel. PPS i forgot! Lizzie does have editing features: https://www.lifein19x19.com/viewtopic.php?p=241240#p241240 they don't work for me right now, but that may be because i have OpenJDK so im gonna try Oracle Java instead PPPS Yes, with oracle8java, Lizzie's [shift+arrow] and delete keys work,... except that
@gogre I agree that in Go you should never take decisions based on the history of the game. So, it seems at first that it is a waste of resources for LeeLa having to drag that 7 move history with it.
From what I've heard, the reason behind it has to do with resolving complicated ko positions. After all, LeeLa should know which moves are legal/illegal , and you can only do that by taking the history of the game in account.
Having said that, for simple review purposes, you could just add the stones in a random order. (if we assume that there are no captured stones)
i want to change my testimony: when it comes to review, history is definitely not bunk, because it tells us how we got into whatever awful mess we are in. you can rewrite/spin history, but you can't change the past.
as to whether anyone would benefit from knowing history insofar as their next move is concerned, i still believe it's only the stones that count, not how they got there, with the exception of kos, for which a history of the entire game is needed, not just the last few moves.
i also think that would be true for a DCNN bot too, but that's not a Lizzie issue.
PS1 recently, i've been loading games into Lizzie and found that she can ponder right away at the end without having to single-shot down the move tree. PS2 and ive learned that left/right does skip sideways, provided that the next variation is at least as long as the one you're in, but i would still prefer it if it skipped to the start of the next variation, and treated variations as if they were in a circle. shift left/right reorders the variations, although it's unclear to me why anyone would want to do that. Ctrl-left skips to the move before a variation; Ctrl-right does the same thing as right,
So here's my suggestion:
up/down steps up/down as at present left/right skips to the start of the previous/next variation spacebar skips to the end of the next variation shift spacebar toggles pondering
Bumping this. I don't know if there's been a change in LZ since it was first raised. I can use a different SGF editor to place stones, save the SGF, then load into Lizzie, and I have no problem getting reasonable analysis results on a position with no history. So it looks as though the principal objection to implementing this is no longer valid?
You're right. It would be nice to allow this. @zsalch do you want to add this? I will add this at least by 1.0 for better analysis / sgf editing.
I have been doing a lot of analysing of my fuseki with LZ lately but a lot of time it gets really obtuse having to play it whole from the first move.
Doing analysis or study with LZ it would really help if we could just place/erase stones without having to play them in order to analyse given position.