News

European Championship 2018 final game (Pavol vs Ilya): review using Golaxy
By Artem Kachanovskyi | European Pros | 23.11.2018 9:52| Views: 6508 | Comments: 2
European Championship 2018 final game (Pavol vs Ilya): review using Golaxy
The final game in the European Championship of 2018 was played on the 5th August between Pavol Lisy 2p and Ilya Shikshin 2p. Pavol won the game and became the new European champion.

At the start of October 2018 Dr Jin Xing and professor You Xiaochuan, on behalf of the Thinker company, provided European professionals the opportunity to use Golaxy for learning purposes for one year.

Here is a short introduction to Golaxy from professor You:

"Golaxy is developed by the Chinese AI company named “Thinker” (深客科技 in Chinese).
Golaxy, while adopting AlphaGo's basic architecture, introduces innovations on model structure, input features, and MCTS simulation methods. It completes model training with less computing resources and training samples than AlphaGo, yet has the potential to achieve the highest possible Go level yet by an AI. Its playing style is easy for human players to understand and to learn. It can also play Go on boards of all sizes and all komi settings. Unlike other Go programs, Golaxy doesn’t compromise even when leading.
Golaxy won the championship of World AI Go Open in the August of 2018."

I guess most of the Go players understand what Golaxy is. It is a highly developed Go-playing AI program, and one of the best in the world. Professor You didn't mention in his introduction one thing that seems interesting to me. Unlike most of the other top AI programs, Golaxy can estimate both the probability of winning, and by how many points one or the other side is leading.

After receiving such a generous gift from professor You, I had the idea of reviewing the final game of the European championship 2018. I already had some experience in using Golaxy, as I had a chance to use it at the Chinese C League 2018. At that time Golaxy was the main coach of the European team.

So, let's get started with the review itself. First of all, here is the game record:
To get an approximate feeling of the game flow, I used Golaxy's mode of uploading the game for processing. Then you can choose which speed Golaxy should use for thinking about each move made in a game: 1 second, 3 seconds or 10 seconds. I chose 1 second. Golaxy estimates winning probability and by how many points black wins after each move, and using this data it creates a chart of the game flow. The red line is the probability of black winning, the blue line shows by how many points black is leading.
Game flowThis chart might be not completely accurate, but it shows approximately what was happening in the game.  It looks like the start of the game was even, but in the first fight White had a better position from move 40 until around move 90. Then Black took the lead and that lead got progessively larger, reaching a maximum of about 20 points at move 155. Then White began to catch up, and around move 220 the game was really close again, but after that Black had a safe lead until the end of the game.

After that I analyzed the game in more detail, move by move. You have to realize that there are a lot of small mistakes even in top pro games.

For example, instead of move 13, Golaxy suggest a more elegant way of playing:

Instead of move 20 Golaxy suggests the following sequence:

These 2 options Golaxy estimates as 0.5 points better than the actual game. Not sure if we can call moves 13 and 20 "mistakes", they might be just not be completely accurate. In this review we will cover only relatively big mistakes and the most interesting points of the game. Go is a very rich game, and sometimes Golaxy has many suggestions which differ from each other, but their effectiveness is similar. That aspect is also something which we can't cover completely.

At move 33 comes the first chance for Black to take the lead. By playing in the following way, he could take a lead of 3.5 points (in the game it's 0):

At move 35 Black could take a lead of 5.5 points (in the game it's -0.5):

Seems like Golaxy doesn't really like the aji that Black has left in the corner in the actual game. As for White, Golaxy also doesn't like this invasion. Instead it suggests creating a corner enclosure in the top right corner (at O17).

With move 45 Black loses 6.5 points (even game ---> -6.5). Black had to play a simple move instead:

Although White didn't find a correct way to punish, Ilya played in his style, with a sharp fighting sequence, but seems like this time it wasn't the most effective way. Instead of fighting a ko with move 50, he could play F11 himself and take a lead of 6.5 points:

Another interesting option for this move was to escalate the ko by increasing its value. By following this choice, White takes the same lead:

Golaxy doesn't really like White's plan of cutting Black, and the game becomes even again.
The following sequence was pretty much forced, and the next mistake Black makes with a light ogeima at move 79. Instead Black could attack White at the left, and it would bring 4 points more (-1.5 ---> +2.5):

Next black falls even more behind with move 81. Golaxy suggests not to cut white and to play the ko. After white reinforces the group on the left, black can cut at D12. That would be 5 points better for black than the game (-6.5 ---> -1.5).

Next it's White who makes a mistake with move 106. It seems like White plays too heavily with the bottom group, not considering a sacrifice at all. With the next sequence White would keep a lead of 4 points:

Note that after Black's move 16, White can still live with his group at the bottom.

After that the situation became worse and worse for White. 

Before move 136 White was already losing by 11 points, but move 136 lets Black take a lead of 17 points. Golaxy doesn't even consider an option of giving up an "eye group" at the left:

So, at the point of move 145 the result of the fight is catastrophic for White, he is losing by approximately 17 points:

Next both sides underestimated the value of capturing the group at E6. Golaxy suggests that instead of each move, for example at move 147 Black could take a lead of 24 points:

Black could do the same at move 159 and start leading by more than 25 points:

I like this plan. Black's center is secure, he has so many points and it is hard for White to close all the potential that they have.

Golaxy doesn't like Black's invasion at move 167. It was still better to take more points and secure the center.

At the time of move 182, White is losing by only 10 points:

The last chance to take cash was at move 199. Black would still lead by 11 points. Indeed, as professor You said, "unlike other Go programs, Golaxy doesn’t compromise even when leading":

After move 206 White has gotten closer, and now there are just 4 points separating the two sides:

With move 220 White could make the game even closer. Golaxy says that it's important to take sente and save the 2 stones in the center. Like this White would be behind by just 1.5 points:

After Black takes 2 stones in the center in sente and connects the ko with move 235, Golaxy says that he is ahead by around 4 points and that White has no chance to catch up:

So, to sum up:
The first half of the game was not bad for White. There were moments when he could take the lead. After getting a bad result in the fight, White tried hard to catch up, and at some point it was really close. But Black managed to keep the lead, and in the end he won the game by 2.5 points.

Here is the photo which shows the atmosphere of the game, right after it was finished:

Congratulations to Pavol for becoming the European champion 2018!

Feel free to comment! Would be interesting to compare with what Leela (or any other AI says).

This article was written by Artem Kachanovskyi

EGF professional Go player.
Born 12th December 1992.
Started to play Go in 1999.
Promoted to 1p in 2016, 2p in 2018, 3p in 2024.
Living in Ukraine, Kyiv.
All his articles on the EGF website.

Comments:
Sinan
#1
25.11.2018 19:47
This is one of the greatest articles ever written, wonderful !
Artem
#2
25.11.2018 20:41
Thanks, Sinan!
I know you like Leela alot, I am glad that you liked a review from Golaxy aswell ;)
Leave a Comment
* Name
* Email (will not be published)
*
Please, enter the name of go in korean!
*
* - Required fields