You did realize this was about Go, right?

Anyway, lately I have been able to defeat Many Faces of Go 12 set at 6 kyu pretty consistently with a 4-stone handicap, but have been demolished each time I tried to step up to its 3-kyu setting. (Actually, I first tried stepping up when I was still taking 6 stones at a 6-kyu setting.) So after finally winning with 9 stones--aided by its early resignation--my plan is to try and slowly reduce the handicap at 3-kyu. (Some of my success at each rank level, no doubt, is owing to a similarity in its corner approaches at each rank. So, no, I don't think that I am a SDK yet!) I would be interested in any comments on my plan! I don't think, however, that this game is worth anyone commenting on. I just feel good about finally winning a game against its 3-kyu setting!

A closer game if anyone wishes to critique my recent play:



Thanks in advance.

I would choose to attack one of white's groups instead. Either K3 or C10

Uh, this is an interesting point to go over. But in this situation it seems right.

Good.

Okay now look at the situation here. Notice how it is different than ?


Also your lower left corner and upper right corner are very vulnerable if white attacks properly. But it seems that you know white will prefer to play B5. Your human opponents may be more creative.
I think there's an article on Sensei's Library. I'll edit it in later.

That's all I have time for now. That shape just stuck out to me because it was something I was told was wrong. Understanding why it's wrong could teach you a lot too.

A few comments on the opening.



It is not too soon for you to play humans, who will make different mistakes.

As for playing agains the computer, may I strongly suggest the 9x9? As a rule computers play the 9x9 better, and have better style on it.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Yeah, I have been fearing that I am training myself to take advantage of Many Faces of Go rather than learning good technique. In particular, changing from the third line opposite corner side pull back (kikari?) after the 3-6 approach to the kick and high corner enclosure is what turned my record around. And MFoG really likes to approach at the edge rather than jump into the corner.

As for humans, I run out of patience waiting for any automatch at KGS. My rank is 20 kyu, and whenever I log-in the vast majority of players are about 4 kyu and up. I can usually get games right away at Tygem, and I need to go back. However, at 18 kyu my opponents seem to wildly and unpredictably fluctuate in rank. (Assuming that it isn't me!)

So, Bill, do you think that my plan of training with MFoG is bad or inefficient? If I understand correctly, the program is crippled at lower levels (not Monte Carlo? smaller library?) but plays solidly at dan level. So, I thought that if I could get to be competitive with 9 stones against its best setting I would be learning good Go rather than how to "game" the program. Of course, I wish there existed a Go club in Northwest Indiana!

Again, I want to thank both of you for your comments.

You have another option now (at your present level). I know it is satisfying to be playing at a lower handicap (and the more open board means more like an even game). But if you are able to play a reasonable game against it at the 3 kyu setting taking four stones you could also try falling back to seven stones against the top (1 dan) level and trying to reduce your handicap from there.

The point is, when MFOG 12 is using its top MCTS level won't be anywhere near the amount of "bad habits". Sure, it will be making overplays compared to how joseki would be played in even games, but that's correct for a high handicap game and the human opponent strong enough to be giving you that much of a handicap would be doing that also.

Hi Mike. Perhaps you misread my statement? Or confused my 6-kyu game with the earlier uploaded 3-kyu game? I finally was able for the first time to play a decent game against a 3-kyu setting by taking 9 stones. My goal is to get that down to 4 stones before jumping to a 1-dan setting. So, I think that your are agreeing with my plan, just overestimating my current skill!

Anyway, thanks for your input.

I think that when my wife goes to the gym today, I will try a 3-kyu setting with a 7-stone handicap and see how close I come.

BTW, I started playing the kick after reading lovelove's recommendation in "How to Become a Dan" (I would link to the section but the L19 page won't load right now; Edit: First comment Lecture 4: viewtopic.php?f=15&t=7273) that the normal move was too slack when playing with a handicap. Of course, I may be misapplying his advice!

Yep, misread it.

BTW, you didn't say what sort of hardware you are running the MFOG 12 on or at what time setting (unlike some of these programs, the strength or the MCTS levels of MFOG aren't absolute --- when the program comes up Fotland has it analyze what it is running on and then time management determines how many playouts can be used to stay withing that).

You should be OK using MFOG 12 at the "3 kyu" level in terms of bad habits as that is also MCTS (won't play the same way every time, won't make the same "mistakes", etc.). But according to the strong players at my club who I had try playing against the program, its strength isn't even across go skills. So the level is an average of somewhat weaker strategic ability with somewhat stronger tactical ability. That would perhaps be like a young child who is a strong player (for his or her age).

I am running my home-built i5-2500 cpu at 3.6 Ghz, 16 GB ram, Windows 7 Home Premium 64-bit. I keep debating whether I should overclock it.

Yeah, I have asked a couple times about when MCTS kicks in. Based on my experience, it seems like MFoG makes a very substantial improvement at 3-kyu. In particular, I don't notice the same repeated pattern of opening moves. Of course, when I tried jumping to 3-kyu before, I was only winning about 50% at 6 stones against the 6-kyu setting. Believe it or not, I never paid any attention to the default time settings--although I did notice it was setting komi to 0, so I now reset it to 6.5 before each game. But checking just now, I see that I have been getting huge time odds! It feels like I play at least 75-80% as fast as the program, though. So, maybe I should try 6-kyu again after adjusting time settings before moving on to 3 kyu. Two steps forward, one step back!

OK, let's deal with this first (I possibly have the advantage of some communication with Fotland). Don't bother overclocking as that won't up the crunch power significantly*. But your CPU is more powerful than Fotland was calibrating for. Even on an older two core 2 GHz machine the 3 kyu level might actually be closer to 2 kyu and with your CPU, definitely so. The 1 dan level probably closer to 2 dan (remember, on an even more powerful machine the program currently is holding a rank of 3 dan). It's not easy to calibrate these programs when they are are adjusting the number of readouts to the hardware and time settings**.

And yes, the "3kyu" level of MFOG 12 is using a MCTS evaluator so you should be OK there. When people warn of the bad habits to be learned playing against the computer that is mostly with the more predictable AI levels of play.


* To advance the MCTS programs two ranks requires perhaps an order of magnitude more crunch power so overclocking 25-50% is no big deal.

** Some of these programs (as distributed, obviously their tournament versions must play by the clock) have an easier time calibrating as they set the number of readouts according to level setting and ignore how much time needed for that on any given hardware. But MFOG 12 as distributed is designed to be able to play with the clock. Thus the level might not come out to be right.

Normal. It is normal not to give komi when taking handicap stones. For now, taking a large handicap (against MFOG 12) leave the komi at 0.5 (I know you said zero but I think this is what you meant, the half point to prevent ties).

While not usual when playing against humans, you might begin giving komi once you have gotten down to a two stone handicap, are winning too high a percentage, but too low a percentage just taking black and no komi.

Thanks for correcting me about this! (And yeah, I was wondering why it kept changing komi back to 0.5. ) I think I will try a setting of 30 minutes apiece. Or do you think that the program play deteriorates more than human players at shorter time controls? (I'm used to computer chess, where computers are much better than human players at quick time controls.) I think that I had it at 60 versus 15 minutes, so it feels like I will have to start over now!

Yeah, I haven't overclocked yet as the cpu is not the bottleneck for my system. I'm waiting for solid-state drive prices to come down and reliability to improve. I have Windows performance scores of 7.5 for cpu, 7.6 for ram, 6.2 for graphics and gaming graphics (one day I will get a second matching video card), but only 5.9 for hard disk data transfer rate (my system disk is a 1 TB WD Black Caviar 7200 rpm SATA II drive; I have some larger, slower drives for storage and back up ). I know that disk speed makes a large difference for programs that access large chess opening and endgame databases. I don't know how much a program like MFoG relies on a database, though.

The key there is that black already has a stone in place on the side star point that prevents white from getting an efficient extension. It's really a sort of tenuki joseki where black played a three-space high pincer, white tenukis, and then black plays the kick to take advantage of the tenuki. Without that pincer stone already in place, black doesn't get enough out of the kick, since the corner is still open to invasion, and white will get a strong settled group very quickly.

Like I said, misapplying his advice!

Well ...... not really, though if you want a more complete explanation contact me off line. I am not saying that that as written searches of large databases don't require a lot of I/O (and so access speed matters). But I am saying that with specialized searches of specialized data it is almost always possible to design the data encoding and database so that "fast search" techniques can be used and only a miniscule percentage of the database is actually accessed (very little I/O actually done). This is the sort of problem that used to be brought to my desk.