Watanabe Can you set everything to "1"? I'm already used to that (laughs). Well, it's okay to be the first move and the 1000th move. If you know it as knowledge.
SugimuraUnderstood (laughs)
From now on, I need a GPU
Watanabe (While looking at the parameter setting screen) I don't know what each of those words mean anymore (laughs). Any shogi players who know this?
Matsumoto You seem to know Mr. Senda (Shota 7th Dan). Now, dlshogi is installed on a machine with great specs. Research seems to be progressing, isn't it?
Watanabe If only I could master it (laughs). Yes, I wonder if I can use it... It's like which one to take. Well, just with the Mizusho, it's completely faster. What's next for Mizusho?
Sugimura There won't be many updates. I feel like I've already narrowed it down. However, the new technology has a lot of improvements, so it seems that it will become stronger and stronger. Now many developers have moved to the DL system. There may be only 1 or 2 teams developing with the conventional system in the strong team.
Watanabe Then next year's tournament...
Sugimura Next year's tournament will almost certainly be the winner of the DL series.
Watanabe Then it won't work on a PC like this (previously with normal specs)?
SugimuraThat's right. Because there is no part (GPU). I can only read 100 moves per second. Still strong, but not suitable for research at all.
Watanabe Then what will everyone do...
Sugimura Everyone will buy that part.
Watanabe A long time ago, what about a PC that was provided during the Den-O battle?
Sugimura At the last Den-O tournament (November 2017) ponanza started using GPUs. It was only at the end that the GPU was installed in the machine provided by Mr. Dospara.
Watanabe Huh. Then everyone buys new.
Sugimura If you buy a really good GPU, it will cost 200,000 or 300,000, but if it's cheaper, 50,000 or 100,000 might be enough.
Watanabe Then it seems like it will be enough to buy it from now on.
Sugimura That's right. It's just going to be tough with a laptop computer that doesn't have a GPU.
Watanabe There aren't many people using notebooks anymore.
Sugimura If so, buy a computer with a GPU. However, when it comes to whether the water master will be completely obsolete, there are times like just now. I think it's quite difficult to make the endgame stronger in a DL series.
Watanabe But in terms of ratings, the water master and the conventional type won't change.
Sugimura That's right. By the next year or so, the rating of DL series is already increasing.
Matsumoto But I don't think there will be many shogi players who only consider DL.
Sugimura I'm not sure about that, but...
Matsumoto Because Mizusho is much easier to use.
Watanabe Well, it's easy to use.
Evaluation value and "winning rate"
Matsumoto Even if you use the DL system, if you don't do it in parallel like before, there may be pitfalls, so it's scary Isn't it?
Sugimura Hmm, that's right. However, to put it the other way around, I found out that Mizusho was completely appropriate for the evaluation in the early stages.
Matsumoto I see.
Watanabe It's about before the pieces collide, isn't it?
SugimuraThat's right. At that stage, dlshogi has an evaluation value of 200. So if Mizusho is "1" or something like that, dlshogi often wins in computer shogi.
Watanabe I wonder if that has something to do with us (laughs)
Matsumoto What do you think? Depending on how you use it, it might be advantageous for early stage research.
Watanabe Could you put it out early? Well, if only I could establish that procedure. From there, I hope I can establish a procedure that I can imitate (laughs).
Matsumoto In the 3rd round of the throne match, it was said that Toyoshima's challenger was good until the middle... I also wrote "Reversal" in the bulletin article, but I reflected on whether it was appropriate or not.
Watanabe Ah, it's about the sealing hand. I think ABEMA's AI had a (win rate) of 60%, but what about that? I mean, it's reversed... But there was a comment saying "Fujii struggled". But it's impossible to know and do it. Hmm, but that. From the second half, there is no feeling of 60% at that level.
Sugimura ABEMA's "SHOGI AI", the evaluation scale is a bit big. I'll put it out from Mizusho.
Matsumoto Yes. I was also lined up with Mizusho, but with Mizusho I didn't get 200 points.
Watanabe What percentage is 50? Correctly. That's not 60%, is it?
Sugimura ABEMA-san, you can't tell me what formula is used to calculate the evaluation value and win rate (bitter smile)
Watanabe I did some research, or rather, I took notes (of the winning percentages) every time I saw various shogi in comparison with this (Mizusho). Then it doesn't fit at all. I was able to come up with a rough guideline along the way, but it didn't match at all depending on the shogi. I thought, "What's going on here?" When I thought about 60% and about 400 points, it didn't fit at all.
Matsumoto Is there something different than just the evaluation value?
Sugimura I think there is almost no doubt about that.
Watanabe Fujii's early research doesn't fit well with everyone. So the software you are using must be different.
Sugimura Fujii-sensei, recently in the first move, I think he definitely refers to ▲96 steps (poke the edge of the left side), but that's dlshogi's recommended move.
Watanabe Ah, that's right. Is that a highly rated tactic?
Sugimura dlshogi always points to ▲96 steps.
Matsumoto I see, it's that kind of era. It is said that the Aii Rook's Three Major Laws are Aikakari, Yagura, and Tsunogawari.
Watanabe Isn't it important to always give your opponent three? Because the preparation is distributed. Nagase, Fujii, and I use three, but Toyoshima-san is the only one who doesn't use three. In the first move, it's almost only a corner change. The type is a little different there.
Matsumoto If you continue to fight the same opponent, will it be difficult to rotate tactics?
Watanabe Hmm... that's right... It's basically like that, but there are types that care about it and types that don't care. Nagase, Toyoshima, and I are the type of people who give orders for tactics on a case-by-case basis. At that station, I brought this today, something like that. It's not like, "For me, this month, this is the theme." 'This station brought this, how is it? ' way. But I think this is the general trend. But in the case of Mr. Fujii, once you start working together, you will continue working together. for several months. That means that the opponent is almost out of the match. Occasionally, the cycle will change, but if you continue to take turns, it will continue, and if you continue to turn around, it will continue. I don't really do things like 'I brought this today'. You can't change by looking at your opponent. I wonder if there is a difference there.
Matsumoto Mr. Watanabe also has the experience of often hitting the same opponents in a short period of time.
Watanabe However, there is no double title match with the same opponent. If it's with a different opponent. If the opponent is different, the battle style is completely different. It's not that much of a problem, but it must be tough to hit the same opponent this many times. Like the old Hanyu vs. Sato Yasumitsu match (23 games in 2005).
Matsumoto The sealer in the 3rd round of the throne match. Mizusho is △9 four steps. ABEMA was △ 5 Shigin.
Watanabe ABEMA gave 60%, and when we looked at it, we thought, ah, is that so? The back ball is bad form, isn't it? In a pattern where humans are most likely to make bad moves in ball flight approach. My feeling is that even if the second move has a good evaluation value of about 200, it won't be able to handle it well. Because there is something like ▲ 4 two steps.
Matsumoto The focal point is Ayumu.
Watanabe When you get hit, you always have three options. There are more ways to escape, but it's pretty tough. Mistakes for this ▲ 4 two steps or ▲ 3 three steps are usually the second pattern. In the main score, the evaluation values are about equal, but it's always hard to win. Because it's always bad anyway. It would be nice if somehow I could get rid of the bad form with a feeling like △ 8 Ippei, but it's quite...
Matsumoto I see.
Watanabe The software says △5 Hachigin is fine, but you don't know if it's good unless you go a long way. Humans can only read up to 10 moves at most, and if they say, ``It's not clear,'' they'll dismiss it, but after reading about 20 moves, 500 (points) is a human being. But I understand. But that's a little Even Mr. Fujii thinks it is difficult to read that far.
Matsumoto Mizusho's evaluation value is 500, and when I write articles, I express it with a nuance of "a little better".
Watanabe Recent software 500 or something like that makes us think, "Eh? "like this. "Why is this 500 away? like. Even humans can tell if it's 600 or 700. If you think you're even when you're doing it, when you get home, it's 500, it's Zara. "Eh, is this 500? like. 500 is the current evaluation value, I wonder if it's about equal for humans.
Matsumoto I see. △ 5 Hachigin who hits a rook in the first move. Isn't this the first move you'll notice?
Watanabe Yes, yes. After △5 Hachigin ▲4 Hachihi, people read △4 Nanabusei. If you do that, there is almost no benefit to being behind. This is a bit risky if you don't know, so I can't point it out. It's pretty competitive. (It's good to research beforehand) If I knew, I could point it out (laughs). But this is not really the scope of my research. I'm a maniac doing this research. There are many other issues to be studied. Moreover, with the current schedule of the two of them, there is almost no time for research at home, right?
Matsumoto I see. The spectators, especially us amateurs, can't help but think, ``Is Fujii winning from a come-from-behind victory?
Watanabe Ah, yes (bitter smile). But after all, the experience is different from that of the opponent, isn't it? Also, people who don't usually use software have a completely different sense of what 60% means. We know that 60% will be given to Zara. Then, how much humans can win from there, I know it's impossible.
Matsumoto On the morning of the second day, Teshima had a winning percentage of 70% at ABEMA, but it was just that moment.
Watanabe Did you get 70%?
Matsumoto It appeared.
Watanabe Hmmm, but the players shouldn't feel that way. Even if you do well on the second move, it will still be difficult to win.
Matsumoto Immediately after the end of the match, a spectator commented, ``It would have been better if △ 4 five silvers appeared,'' and Toyoshima's challenger was surprised.
Watanabe Well, I guess so. But it would be better if the reporter said, "It was good to be on the second with △ 4 five silver." I don't know if it's a consideration of the waiting room or something. Sometimes (not the best) it's completely different. Sometimes it's the teacher's opinion on the spot.
Matsumoto So, recently, during the impression competition, they ask, "Is this a good move?"
Watanabe Yes. When I ask, "Is that true?", some teachers say, "Because the AI said it. △4 We really don't know if Gogin is a soft move or a move that the teacher just came up with. After all, if you can understand that, isn't that player strong (laughs)? I don't know, so I'm seeing off in the actual battle.
Matsumoto △5 Hachigin ▲4 Hachihi △4 Gogin is enough to surprise the Teshima challenger, so it must be a very difficult move to see.
Watanabe Then why is this difficult to see? ▲ 6 I read the move (counterattack from the first move) like Sangin, but it looks more severe than the actual evaluation, this is it. I can't make the right decisions because I'm attacked in close quarters (by the bad form of Tamahi).
Matsumoto I see.
