#101 · Arek Kulczycki
2 Nov 2018
laz, this way is black win, white has to be smart, maybe like this
Public forum discussion.
2 Nov 2018
laz, this way is black win, white has to be smart, maybe like this
2 Nov 2018
Both of you should write a book or tutorial for 2000+ wannabes :)
2 Nov 2018
The mega tournement has begun! Good luck to all! :)
9 Nov 2018
Leela, how do you compute probabilities of winning for each variant? Is it based in same way on Monte-Carlo?
10 Nov 2018
Yes probabilities come from the MCTS algorithm, but where the leaf node probabilities come from the neural net (instead of random playouts).
14 Nov 2018
Leela - could You comment on estimated probability of winning after each move?
14 Nov 2018
@force majeure
You mean in the game messages after each move? That would be just rude! :D
14 Nov 2018
Yeah that was my intention. I think it would Have great learning aspect
14 Nov 2018
I agree, it would be quite rude. At least let me pretend I have a chance. ;)
14 Nov 2018
So this feature shall be activated by messaging leela „teach me” or „stop pretending” :)
14 Nov 2018
Maybe when the game has practically ended, it can post a message with analysis of whole game.
14 Nov 2018
Force, if you've questions on your games, ask me, It's always a pleasure to criticize games of others! :P
In your game vs leela, "obviously" 5 e9 is questionable and it probably gives away the game...
14 Nov 2018
Force, basically, when you play a stone, you should make sure it's connected both sides :D Eheheh
15 Nov 2018
This game? I disagree, 5 e9 seems like an OK choice. And I'm not sure if I like leela's response... I would normally be hesitant to play 5 e9, but if I was told 6 g9 would be the response, I'd be more likely to choose 5 e9.
What move would you have preferred instead? (Also, we should probably stop talking about this game because it's still ongoing...)
15 Nov 2018
shalev, why not 5 b5? I think the position is very hard for black because c2 is just not that strong...
15 Nov 2018
Here is my modest proposal. It's nothing special but maybe it's simple and it leads to decent game for black?
15 Nov 2018
shalev, in general d10 is a counter to e9, so if you play e9 after d10, well, the opponent has already played the counter.
it's like playing this, it's likely nonsense (and i've even experimented with nonsense like this)
15 Nov 2018
About leela response, you need to respect opponent! If it decided for that choice, it knows what it's doing! Probably It has already calculated everything!
15 Nov 2018
It's just hilarious.
15 Nov 2018
Shalev, Arek also has accidentally started a game with leela. So far he is probably winning, but he need to keep his win to the end now.
15 Nov 2018
Lazyplayer, I would never play d10 as a counter to e9 if the k10 is on the board, as in this game.
As for your proposal... it's interesting. I'm usually scared of giving my opponent 3 plays in a row on the short diagonal (at least on small boards), but here you're giving white 4 plays on the short diagonal while playing 4 stones near the acute corners. I would not be so brave.
15 Nov 2018
15 Nov 2018
I agree with shalev on 5.e9 being a very decent move. Reversed logic, like shalev has said, d10 vs e9 is a terrible response given k10
15 Nov 2018
Force actually is imitating leela vs leela, e9 is a very bot-like attempt
But well, imitating leela vs leela may be a very bad approach because leela has done all it's training vs leela hahaha
15 Nov 2018
I try to get him psychologically! His winning rate predictions will crumble!
15 Nov 2018
hahahaha - gzero allows a string with "penny" to report its skewed probability back to opponent - I'll add support for "trash can" too just in case!!
15 Nov 2018
Arek, my evaluation was 99% win for black before b11-b12, now it's only 75% win for black.
15 Nov 2018
You've lost the potential d11-e10 escape for 11th row for... and potential b12-c11 escape for 12th row... for nothing tangible...
15 Nov 2018
And black b10-c10 sequence for helping you connect to top, frankly, not even 75% black win, I'll say, 50%/50% now... and I'm hoping for black to win... Arek, you need to play better!
> Reversed logic, like shalev has said, d10 vs e9 is a terrible response given k10
It's not terrible enough. It still does something and hence it's still crippling e9. It doesn't take much to turn e9 from winning to losing.
15 Nov 2018
you've to understand, e9 is free and already played, obviously it'll be weaker than a fresh stone, but it's already there. it's basically free.
15 Nov 2018
laz, I can only have one of:
d11-e10 /b12-c11 /d9-e9+b8
needed to sacrifice something to gain something because central play was in favour for white
15 Nov 2018
Arek, I still think it's dubious but not 50/50, maybe 80/100 for you rather than 99/100. Now leela has played some more dubious choices so who knows...
15 Nov 2018
Arek, shalev - I don't get the logic behind "d10 vs e9 is a terrible response given k10". Why is k10 affecting that choice that much?
gzero - how accurate is your algorithm? After 2 moves it shows just 35% for gzero.
15 Nov 2018
Because k10 can escape a fourth row ladder that could develop had white responded to e9 with e10
15 Nov 2018
HappyHippo - lol, that was actually my intention of 5. e9. However, I've tricked myself and for the whole discussion I thought that k10 is in upper right corner :D
15 Nov 2018
@ force - it is accurate in that is what the model thinks... ;)
but yeah, add/subtract 15% and it will make more sense, and towards the end game the probability will be closer to reality.
16 Nov 2018
Force, my point is that k10 isn't enough to save e9 from d10. Given we're 2 strong players vs 1 on this, take this with a grain of salt. Maybe bots can tell us their opinions.
19 Nov 2018
It seems that leela will easily remain invincible in the tournament :/
19 Nov 2018
Force, it has lost a game to gzero, but it played a really crazy start...
19 Nov 2018
ahahaha, this obviously does not work :D
19 Nov 2018
Maybe not THAT obviously, but I'm 95% sure it's nonsense.
I wanted this to be the simplest refutation, but it's not verified
14 Mar 2019
After 4 months this epic tournament has just finished! Thanks for the participation!
15 Mar 2019
Good (and tough) games have been played!
Looking forward to the next one... ;)
16 Mar 2019
Actually it has not finished yet, but congrats to Daniel for the amazing performance!
Congrats to gzero and leela too, I will get you next time!
16 Mar 2019
What do you mean "it hasn't finished"? I see all games finished. Will there be a second round?
I assumed we were talking about this tournament: Tournament table [ut.hex.1112.1.1]
16 Mar 2019
Oh, sorry, my bad :)
21 Mar 2019
There isn't a second round - unfortunately it doesn't seem possible anymore to create multi-round user tournaments as far as I can see :(
That said, I'd be glad to manually start a second one if there's interest - and once we all decide on some house rules regarding whether we want to let bots into this one!
21 Mar 2019
Leela is okay with me. What other bots are there?
21 Mar 2019
Well I'm assuming gzero_bot was a bot, wasn't it?
29 Mar 2019
I would play one round tournament as well