It is currently Thu Mar 28, 2024 8:49 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 118 posts ]  Go to page Previous  1, 2, 3, 4, 5, 6  Next
Author Message
Offline
 Post subject: Re: DH records
Post #81 Posted: Wed Jan 01, 2020 7:09 am 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d
Unbelievably, I won the London Open!

Here are my game records. I've only added a few comments and haven't checked with AI yet. My memory is not so good anymore so I have recorded the ko fights and endgame in the wrong order.







Attachments:
2019 London Open 1.sgf [5.27 KiB]
Downloaded 1165 times
2019 London Open 3.sgf [2.99 KiB]
Downloaded 1148 times
2019 London Open 2.sgf [5.19 KiB]
Downloaded 1145 times

This post by dhu163 was liked by 3 people: Bonobo, Elom0, xela
Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #82 Posted: Wed Jan 01, 2020 7:11 am 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d






Attachments:
2019 London Open 6.sgf [3.03 KiB]
Downloaded 1164 times
2019 London Open 5.sgf [3.26 KiB]
Downloaded 1144 times
2019 London Open 4.sgf [2.48 KiB]
Downloaded 1162 times

This post by dhu163 was liked by: Elom0
Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #83 Posted: Wed Jan 01, 2020 7:12 am 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d


Attachments:
2019 London Open 7.sgf [2.31 KiB]
Downloaded 1150 times

This post by dhu163 was liked by: Elom0
Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #84 Posted: Wed Jan 01, 2020 9:11 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
The 89 "probably better" vs Podpera was the one I was talking about (as I think centre access more important than one stone half-captured). But on further inspection even better is j6!

Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #85 Posted: Mon Jan 20, 2020 1:18 pm 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d
I went to the 3rd European Grand Prix Finale in Leksand, Sweden last week, having been told I qualified with my London Open result, despite never hearing about it before. I was excited to meet and play against European pros for the first time even if I expected to lose all my games.

However it seems my results were the talk of the tournament as no-one had heard of me before and it didn't seem to fit my 4d rating. Beating two European pros surely trumps winning the London Open! It is unbelievable I finished 4th/16 in a field of 6d-3p.

Game 2 is the only game that wasn't live-streamed, and I've added a few comments and variations overall.

Games 3 and 5 were reviewed on Twitch https://www.twitch.tv/europeangofederation







Attachments:
2019 Grand Prix 3.sgf [30.72 KiB]
Downloaded 1087 times
2019 Grand Prix 2.sgf [3.7 KiB]
Downloaded 1096 times
2019 Grand Prix 1.sgf [13.77 KiB]
Downloaded 1092 times


Last edited by dhu163 on Mon Jan 20, 2020 1:21 pm, edited 1 time in total.

This post by dhu163 was liked by 4 people: dfan, Elom0, Uberdude, xela
Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #86 Posted: Mon Jan 20, 2020 1:21 pm 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d






20221115

just starting to notice strange patterns particularly in game vs SF.

chat says f to support SF, and my stones form F shape. Shapes look quite rude actually given coordinates. NoPo? Perhaps snooker shapes?Pavol was recording.

curious position of GL, GF, GG, HI, ko for PP?

I think this is a game that some higher power (than me at least) was using to make a pun about the player's names. Not sure if there are other sides to it.

I remember saying something like "I was less scared of Pavol because I saw a game he played against Andrew Simons. He was behind the whole game until the end when he started messing around in the corners. I had to do something like that."

"I don't know why the AI keeps improving." "Maybe it doesn't" "It does, ..." (etc?)

"I could show you my game against Stanislaw."

"Is that really punishment?"

"DH: I've already eaten, I'm just here for the ...? SF: atmosphere? DH: I guess?"

"There was a BBC article which said that an organ donor's DNA spread to the rest of the body." But I can't find it anymore.

plastic bags, ...
"Therefore Poland is the best."

"4 seasons, mushrooms, mozzarella, pepperoni, anchovies." "not quite."


20221125

Looks like morse code in the middle. Something to do with circuit diagrams. I suspect isfet. Note that huai4 is Mandarin for bad.


Attachments:
2019 Grand Prix 6.sgf [23.91 KiB]
Downloaded 1085 times
2019 Grand Prix 5.sgf [6.8 KiB]
Downloaded 1083 times
2019 Grand Prix 4.sgf [10.82 KiB]
Downloaded 1070 times


Last edited by dhu163 on Thu Nov 24, 2022 8:09 pm, edited 9 times in total.

This post by dhu163 was liked by: Elom0
Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #87 Posted: Sun Sep 20, 2020 2:58 pm 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d
I won the corona masters league B of 5d european players, but have lost all my league A games (EGF academy) so far. not so great.
I just played katago on my home PC, with 20 playouts per second and 10 seconds per move. I won a 5 stone handicap game where I was scared for the life and death of a big group of mine for most of the game, trying to make eyes while being chased into the centre, but analysis says I was leading by 30 points and eventually kg overplayed trying to kill after I made a ko for life and it ignored my ko threat and I killed a big surrounding group. Then I wanted to try an even game, below. lost by 18.5 points (I can win the last point ko) which isn't so bad. :)

This is somewhat surprising considering my long losing streak (20 games ish) to bots on kgs. but its probably because the game entered endgame very quickly with no fighting, and the low numbers of playouts. The majority of the margin was lost in the endgame. I'll probably try again more in future



it says leelazero, but I changed the engine to katago 1.4.5 40 block


Attachments:
katago training 1.sgf [107.97 KiB]
Downloaded 981 times
Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #88 Posted: Sun Sep 20, 2020 6:31 pm 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d
D8 was a move i've seen AI play, but in this case kg said it wasn't good and C9 or E9 were good. if E9, i know that black normally has to sacrifice if white cuts even with the big wall, even if the stones were one space closer to the wall, but it might be marginally different when the cutting point is at F4 rather than E4.

D8 was played by AI when black's wall was just with E4 extend and white keima F2. The key is to defend the central weakness with one move and not give the opponent ways to make shape while reducing black's area whether in the centre or side. It is more a move to make life cleanly.

presumably its overconcentrated in this game and hence a bit too gote. This maybe means both that it was more valuable for black to spend more moves to secure the territory and also that if white got moves to reduce (the centre vs C9 or the cut if E9) then black at least didn't have to worry about the safety of the group. In the game it felt noticeably worse than either C9 or E9 especially when I connected up to the upper left and when white clamped in the lower left, but perhaps that is just because KG played to counter it and make it look bad. D8 itself lost only a little from the analysis. In conclusion, I suspect that D8 looks good if black can get the B3 hane or something like that, or if the C12 group is weak.

I found it interesting how powerful G3 is against the white wall on the right, even though KG didn't suggest it in this position, it didn't lose points either. my direction in the upper right however was judged incorrect and I should have taken sente blocking on the other side to attack with the cut in the lower right, since I had already set it up with G3.

Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #89 Posted: Mon Sep 21, 2020 5:04 am 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d


I won a 2 stone game, but the AI played so many bizarre moves that aren't at all suggested by KG or leelazero, so I wonder if I've somehow messed up the settings and it is a weak leelazero playing in disguise? Does anyone know how the lizzie AI works?


Attachments:
katago training 2.sgf [7.89 KiB]
Downloaded 938 times
Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #90 Posted: Mon Sep 21, 2020 11:51 am 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d
I've just tried out sabaki, but it doesn't seem to allow me to play against AI. When i do F5 as in the tutorial, it is always engine vs engine even when I set the players. furthermore when I ask the engine to make a move, it can play on where a stone has already been played or other bizarre moves, and most of the time plays the move in the position two moves before the current leading to many new branches of a game ...

so I thought I would use lizzie still, with suggestions turned off, but using analysis mode so that I am sure it is katago. I gave kg 10-30 seconds at 20 playouts/seconds

so vs katago 1.4.5 40s509 (40 block) for real this time. I lost by 11.5 in a 2 stone game. It felt worse than that, but I guess I lived with my main groups. I didn't do a good job in the opening, which felt like my main problem this game. my supposed attack ending up leaving all my groups weak

also, I finally won a league A game, killing a big group in my opponents area, when they had played too many (4 ish) 2nd line moves to reduce my moyo.



Attachments:
katago training 3.sgf [351.6 KiB]
Downloaded 1002 times
Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #91 Posted: Thu Sep 24, 2020 5:52 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Glad to learn you are in the EGF academy. :tmbup:

Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #92 Posted: Wed Oct 21, 2020 7:53 am 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d
back in PETC
won my first game comfortably, added my short AI-less review, and also AI review which mostly supported some of my opinions on the mistakes we both made.



Attachments:
Pandanet 20-21 ETC 1.sgf [17.4 KiB]
Downloaded 872 times

This post by dhu163 was liked by: Uberdude
Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #93 Posted: Wed Nov 18, 2020 5:05 pm 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d
So far, I've only added (lots of) kg's variations in the opening at around 1k playouts



edit (20201121): updated sgf with variations at 100k playouts. I found bteuber's google cloud shell code to run leelaz on cloud gpu. It works very nicely and I even added my own code to make it work with the latest katago (on my github dhu163). It costs around £0.70 per hour.


Attachments:
Pandanet 20-21 ETC 2.sgf [38.4 KiB]
Downloaded 810 times
Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #94 Posted: Wed Dec 09, 2020 5:33 am 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d


the most solid game I've had so far. I didn't attack while's weak centre group so well and made a mistake early on, but overall it felt like I had a solid control of the game the whole way though.

I thought I gained in endgame, but apparently I lost points ...

edit(20 Dec: in trying to understand how I lost 10 points in the endgame, I added lots of AI variations. It seems obvious in hindsight.).


Attachments:
Pandanet 20-21 ETC 3.sgf [40.09 KiB]
Downloaded 749 times

This post by dhu163 was liked by: Uberdude
Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #95 Posted: Wed Jan 20, 2021 10:01 am 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d
A complete mess of a game. From the AI a good proportion of the first 100 moves lose 5 points or more. Partially this is because it was a complicated fight over big areas and groups, but also poor play from both sides.
I started with a joseki I was familiar with and my opponent fell into my trap and I was 4 points ahead at move 47. I proceeded to play completely the wrong refutation and got 4 points behind from move 48. I thought B's shape was so bad on the right I was shocked he didn't defend but when I attacked I got into trouble, missing a shape point at the middle of 3 stones to defend my shape to allow capturing his cutting stones and I thought I had a winning chance. My tight kosumi in the game simply got captured, I couldn't capture the cutting stones and my outside was low on liberties. Despite taking over his lower right 4-4, it was a mess.
I felt behind the whole game from this point but apparently it wasn't so bad and I was ahead a good deal for most of the game until the very end when I shouldn't have tried to kill everything in his lower right. I was ahead by 24 points at the end since my opponent totally misplayed the 1000 year ko wasting ko threats and giving me more threats. Note this AI analysis does contradict much of my post game comments on the britgo website.



Attachments:
Pandanet 20-21 ETC 4.sgf [25.34 KiB]
Downloaded 741 times

This post by dhu163 was liked by: Uberdude
Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #96 Posted: Thu Jan 21, 2021 1:02 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
I was surprised you didn't cut outside for 34 or later as I thought that was the joseki, but presumed it was some newer variation than my knowledge that you had researched.

Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #97 Posted: Thu Jan 21, 2021 4:18 am 
Lives with ko

Posts: 144
Liked others: 134
Was liked: 23
>I trained hard for this game, expecting Rob Van Zeist....

Out of interest, what kind of training do you do (if you don't mind giving away your pre-match secrets?).

I watched this game live and it looked really chaotic, and White in particular looked incredibly busy trying to patch up/defend whilst trying to take points. As a spectator, I enjoyed the way you did a good job of keeping it all the plates spinning, seeing the game through to the endgame, and unleashing your longstanding devastating aji.

Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #98 Posted: Thu Jan 21, 2021 1:45 pm 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d
Quote:
I was surprised you didn't cut outside for 34 or later as I thought that was the joseki, but presumed it was some newer variation than my knowledge that you had researched.

some variations already in the file. but yes, I have researched it somewhat. I know the first few key moves but not what next. So I knew I was playing the right moves up to that point. Still, I've lost several times playing that joseki both against humans and AI. My only wins were the matches vs Tanguy, Lucas and now Geert. And even in those games, every time I got a local loss.

Quote:
>I trained hard for this game, expecting Rob Van Zeist....

Out of interest, what kind of training do you do


L&D: goproblems.com, EGF academy. I've worked on 101weiqi's daily set pretty much every day.
Playing against KG 30-90 moves (i.e. until I get 10 points behind)
Analyse other PETC games with AI. I tend to go very slowly so have only gone through 1.5 so far.
Play go online

From my AI analysis of this game, KG's suggested sabaki with my weak group had several miraculous moves, based on the fact B's cutting groups were very heavy concerning the ko, especially those revolving around the K11 cut. There's a lot to learn from it. It already wanted to tenuki when I was scared of death every move from move 108. Of course I was also scared of death in the lower right ...
My play on the lower side was pretty shoddy too as the carpenter's square looks close to death. I regretted not playing M2 instead almost immediately.

Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #99 Posted: Wed Mar 10, 2021 3:40 pm 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d
"Surprisingly relaxing win". A quick win by annihilating my opp Matias Pankoke. It felt like my opp played very badly playing tight moves against my thick groups that were easy to refute. But it turns out that with AI analysis, I played pretty well too, including move 108 that the AI disagreed with saying it was worse by 2 points until 200k playouts after which it said it was the best by 0.7 points. I am pretty pleased to be told that move 72 was perfect locally as it wasn't my first instinct.

I added up the the 12 times out of 55 that my move wasn't KG's top move, summing the winning percentage and score discrepancy with the top move, and got:

WP 68.2 Score 7.7

This must be a personal record by far in terms of minimising mistakes. I'm sure that on a normal game of mine, my score discrepancy would be 50-400 points. Even my last game against this opponent (I am 3-0 up now), I was winning by 25 points in the end but was 80 points up at one point and missed many chances to save a group.



It seems we misset komi

In other news I am playing in a casual online league.
Ho Yeung Woo: Win: I started with a poor opening from me and aggressive play from my opp. It was a tough fight in the centre as he was so thick. However, when he submissively made life on the second line, I killed lots of cutting stones in the centre with my weak group. I was leading by 20, but my endgame was terrible and won by 5.5
Elian Grigoriu: Win: I researched a joseki so gained in the opening, both missed some big moves in a corner, played some perfect sabaki sequence against his 3-3 moyo, and perfect defense just backing off against his 3-3 invasion. I played softly when he attacked my sabaki group, letting him connect when apparently I should have counter-attacked already and tried to save everything. He tried to use his weak group to attack one of my groups and I killed one of them on a large scale. Leading by 80 points, I skillfully made life for my only other weak group 2, but perhaps due to lack of sleep, missed how he short my liberties, threatening to save cutting stones. I made a seond eye with the group he was attacking that was already in a one eye kills no eye situation. He killed my big group 2. I won by 7.5 points, losing another 10 points in endgame.
Matias Pankoke: Win: I gained a lot in the opening after an unreasonable cut and when he took away an eye of my wall in gote, but played too softly just connecting my stones so all his weak groups lived and I was only ahead by 10 points at move 40 or so. He continued to lose points in the corners in order to reduce my wall, including when I played a 3-3 invasion into a shimari where I only had ko for life, in order to use my thickness and to try to attack the shimari. We both played pretty solidly, I increased my lead to 20 points. He tenukied when I threatened that shimari group and I killed it in a way where my surrounding groups where unconditionally alive. He cut off a big group of mine in return, but the AI repeatedly said I could make life. For 20 moves I missed tesujis, and failed to live. The score bar would jump 40 points every move we played. I was ahead by around 25 at the end due to poor endgame from me and my dead group.
Lucas Neirynck: Win: I was leading the whole game almost after a weird joseki choice. There was only one move where I overplayed and suddenly he had a ko for the entire lower side and corner that I couldn't remove, even though I could tenuki and it would still be ko. He would have been 20 points ahead but missed it and my group was unconditionally alive. My highest lead was around 10 points, but it gradually disappeared somehow when I failed to manufacture big enough ko threats for a ko for my corner's life. I won by 2.5.
Valerii Krushelnitskyi: Loss: The game was close the whole way through, mostly within +/-5 points. I messed up a joseki in the first few moves but he messed up the follow up to the refutation. I underestimated a cutting point where he was very thick and missed a cut of his at a critical point. My defence of my own side territories was fairly flawed and in the end, missed a chance to grow my area by attacking his group. I failed to kill his reduction and lost by 5.5.

I used to be proud of my endgame, but the evidence lately is that my endgame has become relatively full of mistakes.


Attachments:
Pandanet 20-21 ETC 6.sgf [37.66 KiB]
Downloaded 944 times
Top
 Profile  
 
Offline
 Post subject: Re: DH records
Post #100 Posted: Thu Mar 11, 2021 6:35 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Nicely done Daniel :tmbup:

Re his 35 hane rather than jump, hane was also my instinct. Although I know of the clamp, it's not a natural move for me.

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 118 posts ]  Go to page Previous  1, 2, 3, 4, 5, 6  Next

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
cron
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group