Class Reunion - Class of 1969 - Liberty, Mississippi
Title: Crazy Stone Deep Learning -The First Edition-
Genre: Casual, Strategy
UNBALANCE Corporation, Rémi Coulom SAS
Release Date: 25 May, 2016
I like to review my chess games with a chess engine.
I would have liked to review my go games with a go engine in the past.
Now I do!
This is the first available go engine for this task. And I love it, it is a new era.
(Cons: Publisher does not offer upgrade from previous versions. Bought it anyways, because this is my most valuable go tool. Thanks Remi Coulom!). I love it. This is a nice board game and great AI inside.. Human like play. Positional calculation is a major plus. But the program has a weakness with ladders.. Incredible! I've been playing Weiqi for a while before getting this software, and now play at about 4-kyu, and I'm greatly impressed by the AI of Crazy Stone Deep Learning. I usually play against the 7d computer level on 19x19 when I have a lot of time. 6d or 5d when I'm in the mood for a quick game. 3k to 1d when I'm feeling like trying to beat a stronger opponent. The higher levels teach me a lot about the game. Mistakes at my level are quickly punished by the AI. I often use the [estimate] Territory function when experimenting against stronger levels so I can see what works and what doesn't, and why. The interface, music and sound effects are simple but quite comfortable. I look forward to The Second Edition.
Of course, I also do some study from books and videos online to make best use of my "Go" time, so I can improve faster and enjoy playing this and stronger human players even more.
Very well worth the money if you're serious about improving your game!! It's rare I can ever play a 1d+ human player in my small city, so being able to play 7d anytime as often as I like is huge for me. Thank you.
If you're a weak player, you might not like the price, and might want to buy a $10 book first, or buy an easy game instead if you're not serious about Weiqi \/ Baduk \/ Igo \/ "Go". However, I wouldn't discourage a beginner from buying this if serious about the game and can afford the investment. You can play against the 13 kyu level AI on 9x9, 13x13 and 19x19 boards, and work your way up the levels. (A beginner is 30-kyu but usually improves quickly.)
If you're quite strong and don't want a $20 program you can beat easily, this is well worth $80.
For those mentioning AlphaGo, you can't buy it, and it runs on a neural network of something like 2000 cores, for a super corporation's AI research and development that will someday "go" way beyond board games. So, why compare AlphaGo to $80 software you can enjoy right now?. I have one question, it is so different between playing white and black ? Black plays worse than white.. Ok, so before everything else, a little about me: I'm not really strong at go, I played for two years and reached 12~13k on KGS, then I had to stop for two years to focus on studying. When I got back into go 2 months ago, I decided to train hard in order to "get my old 'strength' back". And that included playing against this bot once it was out.
This bot beats me everytime (of course), but it doesn't feel at all like I'm playing a 7d. My first game against it (unlimited time, 7d strength) was lost by 40.5 points in an even game (in case you're interested, here's the link: https:\/\/gokibitz.com\/kifu\/VyqbrOUmb<\/a> ). This is a bot that was supposed to give me 17 stones at the time, and it only beat me by 40.5 points.
In case you're wondering, no, I'm not 3d, I'm around 8k on OGS. So, either it isn't 7d \/at all\/, or it played slack moves once it knew it was ahead (like AlphaGo, maybe). I personnally think it's a mix of the two. On your everyday computer, I don't think it reaches 7d, unless you force it to do 10M playouts thanks to the analyze feature, but that would mean a looooot of time for each move.
The interface is also extremely old school. While it isn't ugly, it lacks game setting customization: you can only choose between predetermined main times and byo-yomis, the komi can't exceed 9.5 (so no reverse-komi handicap games), and you can only either choose between a set time for the bot to use or an unlimited time and setting a rank, so no option to "do at least X playouts per move). Funnily enough, if you put it on the "unlimited time, 7d rank", it almost never thinks about moves for more than a few seconds, but when you set it with a limited time per move, it thinks a lot more and usually plays "better" (as far as I can observe at my level, anyway).
All in all, it is very expensive, and only people who are really serious (and can afford the price) about learning go will even want to consider buying this.
However, I would still recommend this to those because of the analyze feature I mentionned earlier, you can make it analyze a position for as long as you want, and this way you can review games you played, see what moves it considers bad, and what it proposes instead. Playing games against it has also allowed me to progress a lot (because, at my level, it still is way better than me).
If you're patient, I'm sure a free strong enough engine with similar features will be out within 1~2 years, and by the time, (hopefully) CrazyStone Deep Learning version 2 with more customizable game settings and an even stronger engine.
I personnally hope V2 will be free for those who purshased the V1. In the meantime, I need to practice more :)
EDIT: I've run a series of games on OGS to test its strength, it settled at 2~3d with 10 seconds per move and high priority (on a i5-3210M @ 2.5GHz), so it is decently strong even on an average computer.. There's a user-defined tag for this product that reads "casual." I can't really say that this product is for the most casual observer of Go\/Igo\/Weiqi\/Baduk.
However, what I can say is that if you're an amateur player looking for an AI Go sparring partner then here's your option on Steam. In fact, it appears to be one of the most compelling options at the moment for playing this game on Steam (the other being Tabletop Simulator if you have a friend).
Verdict: If you want to Go it solo on Steam (all easy puns intended) and practice against one of the more competent AI's on the market, here is your choice. But for $80 let's hope that future updates or "The Second Edition" will be included.
I'll recommend it for the purpose of hopefully inspiring more products such as this to hit the Steam Store. With even more hope, I'd like to see something like KGS or Tygem on here as well.. First things first \u2013 this game will not teach you the rules of Go. If you\u2019re a total beginner, this isn\u2019t the game for you: there\u2019s no tutorial, and the documentation is very basic, so if you\u2019ve never played Go before, please check out some of the excellent resources available on the web first. (One excellent place to start would be Reddit at \/r\/baduk \u2013 Baduk being the Korean name for the game.)
If you already know how to play, then the question you should ask is, is this game for you?
The obvious thing the game offers you is the AI. Rated up to 7d, this is the most powerful commercially available Go AI, and can challenge all but the most skilled amateurs. Its minimum setting of 13k, on the other hand, should match up pretty well against anyone with more than a handful of games\u2019 experience. In addition, the game supports handicaps up to 9 stones, and based on a couple of games, it does a good job of playing handicap Go (far from universal amongst Go AIs).
So while a player of almost any skill level can get a good game from Crazy Stone, all but the strongest could do the same against any number of cheaper or free AIs available. Given that, why would you pay such a steep price for the best AI?
The reason is the item at the bottom of the features list: the Analysis Engine. Crazy Stone allows you to review and analyze games you\u2019ve played against the computer as well as games from the major online Go servers (anything that can export a game record in standardized format). For each move you can see Crazy Stone\u2019s analysis \u2013 including the top 20 moves it considered and the relative value of each. If you\u2019re willing to put in the time to go through it, this is almost like being able to play a teaching game any time you want, having a 7d grade your moves.
For me, the Analysis Engine is worth the entire price of admission. I\u2019m roughly 1k-1d strength, so I usually know most of the moves that Crazy Stones evaluates, but going through the analysis gives me tons of food for thought about the differences between them.
Technical notes: this program can use all the CPU you can give it. I\u2019m running this on a fairly beefy CPU i7-5930K, 6-core 3.5Ghz \u2013 looking at the Analysis Engine, this allows it to simulate about 10,000 games per second. Generally speaking, it seems to take around 30,000 simulations to get a good analysis for a move, so about three seconds on this machine. I also tried it on an i7-3770K, 4-core 3.5 GHz \u2013 7,000 games\/second, and an i7-4810MQ, 4-core 2.8 GHz laptop \u2013 6,000 games\/second. So it seems like most 4-cores would need around five seconds of analysis time per move. Unfortunately, I don\u2019t have any duo-cores to try out, but you can probably figure on doubling that time to ten seconds.. pricey but worth it
I do a quick blitz game when i have nothing to do or after I lose an online game to cool my head down.,
pretty good if you are a noob because you can set AI to highest lev...