Kirby's Study Journal

Uberdude · **#981**

Shaddy wrote:

YOUR CORNER IS UNSETTLED

But is f1 a clever way to settle it in sente (trying to get d1 sente)?

Kirby · **#982**

Shaddy wrote:

YOUR CORNER IS UNSETTLED

True. If I reflect honestly about this, I did have a feeling that the corner could be in trouble, but I didn't read it out clearly. It feels difficult to give up initiative and to play a move to save it, though.

Maybe Uberdude's idea is a good way to save it in sente.

Kirby · **#983**

Bill Spight wrote:

Kirby wrote:

Some general points to learn from this game:

Let up a little bit on the pressure when the position can be reasonable for me.
Consider alternatives to joseki I play by habit. In this case, maybe pressuring the left stone more would have been better.
Learn from 'TheCaptain': When I have a goal to accomplish, think about solving it indirectly - create a weakness in the opponent's shape first, then exploit it (for example).

I think that's what I should focus on for now.

Nonetheless, I was honored to play a legend today :-)

Reading this, How Not to Play Chess by Znosko-Borovsky came to mind.

Znosko-Borovosky wrote:

Do not create weak points in your game for your enemy to seize.

Do not make the opening moves automatically and without reflection.

Do not be content with attacking an existing weakness; always seek to create others.

Surprisingly similar! :-)

Shaddy · **#984**

It looks working. I wanted to be sure he knew, though

Kirby · **#985**

Inspired by all of the fancy AlphaGo matchups this week, I took another look at the nature paper (which I still haven't fully digested, by the way). I was particularly interested in the input feature selection for the neural networks. As I understand, these are deep neural networks, so learning from the initial feature vector may serve as some sort of first layer, but then subsequent layers, in a way, chunk the information into more complex learned concepts.

Anyway, it seemed informative to me to look at the initial set of features, since these may form a basis of AlphaGo's initial learning - or birth into the world of go, perhaps :blackeye:

The table is listed here:

Feature	# of planes	Description
Stone colour	3	Player stone / opponent stone / empty
Ones	1	A constant plane filled with 1
Turns since	8	How many turns since a move was played
Liberties	8	Number of liberties (empty adjacent points)
Capture size	8	How many opponent stones would be captured
Self-atari size	8	How many of own stones would be captured
Liberties after move	8	Number of liberties after this move is played
Ladder capture	1	Whether a move at this point is a successful ladder capture
Ladder escape	1	Whether a move at this point is a successful ladder escape
Sensibleness	1	Whether a move is legal and does not fill its own eyes
Zeros	1	A constant plane filled with 0
Player color	1	Whether current player is black

For some reason, the bolded entries above having to do with liberties, stick out to me. Capture size, self-atari size, player color, etc., seem somewhat obvious to me, but sometimes knowing the number of liberties for a group is non-intuitive at a glance. Specifically, in many cases, I do not typically consciously count out the number of liberties that a group has. It's more about my feeling than anything else. I get a "feeling" about the count of liberties based on experience. Consciously counting the literal number of liberties is sometimes not useful, since there are corner situations, eye-no-eye situations, and other complexities that make the practical number of liberties different than the "real" number of liberties. Richard Hunter has an interesting book on this, but I haven't internalized all of the rules he lays out.

Most of the time, my process for identifying strength of groups and/or winners of capturing races is either:

Feeling (e.g. "this group feels stronger than that group")
Visualization of a capturing sequence (e.g. "black, white, black, white, ..., ok, black is captured")

Presumably, AlphaGo's interpretation of liberties initially started out from the initial feature - literal number of empty spots around a group. Then, through different layers of the network, maybe it internalized some mechanism of giving a good value to a position, still based partially on the literal liberty count.

From this, I wonder if it is worthwhile to add a new item to my methodology, above, for identifying strong groups and/or winners of capturing races:

3. Count actual empty spaces around a group of stones, understanding that certain circumstances may imply a different result.
Because it seems that AlphaGo would still use the feature in its basic form for evaluating a position. Maybe I should add that to my basic set of inputs as well.

Or maybe this is just one way that computers think differently than humans...

EdLee · **#986**

Hi Kirby,

Quote:

Count actual empty spaces around a group of stones, understanding that certain circumstances may imply a different result.

I think this first sense of Liberty is necessary but insufficient; the key is the second sense of Liberty.

Kirby · **#987**

EdLee wrote:

Hi Kirby,

Quote:

Count actual empty spaces around a group of stones, understanding that certain circumstances may imply a different result.

I think this first sense of Liberty is necessary but insufficient; the key is the second sense of Liberty.

Agreed. What I'm trying to express is that I typically pay attention to the second sense of liberties either by feeling from experience and/or explicitly visualizing, for example, a capturing sequence. I think that this focus on the second sense of liberties has been beneficial.

Up until now, however, I've pretty much completely ignored the first sense of liberties - it didn't matter to me, unless I was explaining someone the rules of the game.

What I'm curious about is if it may be worthwhile to explicitly think about the first sense of liberties as well, since this appears to be an input feature to the neural networks that AlphaGo uses.

Up until now, only thye second sense mattered to me. But I wonder if it's worthwhile reconsidering my views - maybe not, but since it's something AlphaGo uses as input, I'm second guessing myself.

EdLee · **#988**

Hi Kirby,

I agree ( and recently came to similar conclusions, completely independent of DM ). One way to test the theory is experiment and note the results. :mrgreen:

EdLee · **#989**

Hi Bill

No to smoking; drinking in moderation. :mrgreen:

Watching AlphaGo v. pro team, live

Schachus · **#990**

I think this is for enableing lines of thought like: "let me save this sente exchange for later, since it could potentially induce a damezumari on my part later". So moves are considered more carefully, if they reduce the liberties of the group they belong to and I think for that even a straightforward "count the empty spaces around the group after my move" can help to get a picture, whether liberties might get short(remember this network is "just for intuition", reading comes in differently). I doubt this has a more involved liberty count, because it needs to be done in preprocessing, before giving the position to the NN.

Kirby · **#991**

Schachus wrote:

I doubt this has a more involved liberty count, because it needs to be done in preprocessing, before giving the position to the NN.

Yes, me too. My basic idea is simply that AlphaGo considers a basic liberty count at least on some level, whereas I often do not.

Kirby · **#992**

I played another game today.

I lost against KGS 1k, giving two stones.

Here's a review:

Kirby · **#993**

Highlights...

Position 1

Click Here To Show Diagram Code: [go]$$Bc $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . X . . . . . . X O O . | $$ | . . X , . . . . . , . . . X . X X O . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . O . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . , . . . . . , . . . . . , . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . W . | $$ | . . . X . . . . . . . . . . . . O X . | $$ | . . . . . . . . . . . . . . . O . X . | $$ | . . . X . . . . . O . . . . . , . . . | $$ | . . . . . O . . . . . . . . O . X . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

The marked move above is probably too much. That's because, after his cut, etc., the cross-cut seems effective:

Click Here To Show Diagram Code: [go]$$Bc $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . X . . . . . . X O O . | $$ | . . X , . . . . . , . . . X . X X O . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . O . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . , . . . . . , . . . . . , . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . 2 . | $$ | . . . . . . . . . . . . . . . . 1 O . | $$ | . . . X . . . . . . . . . . . . O X . | $$ | . . . . . . . . . . . . . . . O . X . | $$ | . . . X . . . . . O . . . . 5 , . . . | $$ | . . . . . O . . . . . . . . O . X . . | $$ | . . . . . . . . . . . . . 4 3 . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

Instead, I should probably just pull back:

Click Here To Show Diagram Code: [go]$$Bc $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . X . . . . . . X O O . | $$ | . . X , . . . . . , . . . X . X X O . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . O . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . , . . . . . , . . . . . , . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . W . . | $$ | . . . X . . . . . . . . . . . . O X . | $$ | . . . . . . . . . . . . . . . O . X . | $$ | . . . X . . . . . O . . . . . , . . . | $$ | . . . . . O . . . . . . . . O . X . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

Might be OK.

Position 2

Click Here To Show Diagram Code: [go]$$Bc $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . B . . . X . . . . . . X O O . | $$ | . . X , O . . . . , . . . X . X X O . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . O . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . X , . . . . . , . . . . . , . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . O . | $$ | . . . . . . . . . . . . . . . . X O . | $$ | . . . X . . . . . . . . . . . O O X . | $$ | . . . . . . . . . . . . . . . O X X . | $$ | . . . X . . . . . O . . . . O X O O . | $$ | . . . . . O . . . . . . O . O X X X . | $$ | . . . . . . . . . . . . . O X X . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

Not sure where to play after black's marked move. I wanted to try this:

Click Here To Show Diagram Code: [go]$$Wc $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . 2 X 1 . . X . . . . . . X O O . | $$ | . . X , O . . . . , . . . X . X X O . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . 3 . . . . . . . . . . O . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . X , . . . . . , . . . . . , . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . O . | $$ | . . . . . . . . . . . . . . . . X O . | $$ | . . . X . . . . . . . . . . . O O X . | $$ | . . . . . . . . . . . . . . . O X X . | $$ | . . . X . . . . . O . . . . O X O O . | $$ | . . . . . O . . . . . . O . O X X X . | $$ | . . . . . . . . . . . . . O X X . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

But it just seems hard to make much of the group there.

Position 3

Click Here To Show Diagram Code: [go]$$Bc $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . X X X X . . X . . . . . . X O O . | $$ | . . X O O O . . . , . . . X . X X O . | $$ | . X X X O . . . . . . . . . . . . . . | $$ | . O X O . . . . . . . . . . . . O . . | $$ | . . O . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . X W . . . . . , . . . . . , . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . O . | $$ | . . . . . . . . . . . . . . . . X O . | $$ | . . . X . . . . . . . . . . . O O X . | $$ | . . . . . . . . . . . . . . . O X X . | $$ | . . . X . . . . . O . . . . O X O O . | $$ | . . . . . O . . . . . . O . O X X X . | $$ | . . . . . . . . . . . . . O X X . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

Again, not sure about the marked move, above. I want to avoid getting heavy, like this:

Click Here To Show Diagram Code: [go]$$Bc $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . X X X X . . X . . . . . . X O O . | $$ | . . X O O O . . . , . . . X . X X O . | $$ | . X X X O . . . 5 . . . . . . . . . . | $$ | . O X O 2 1 . . . . . . . . . . O . . | $$ | . . O 4 W . . . . . . . . . . . . . . | $$ | . . . 3 . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . X . . . . . . , . . . . . , . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . O . | $$ | . . . . . . . . . . . . . . . . X O . | $$ | . . . X . . . . . . . . . . . O O X . | $$ | . . . . . . . . . . . . . . . O X X . | $$ | . . . X . . . . . O . . . . O X O O . | $$ | . . . . . O . . . . . . O . O X X X . | $$ | . . . . . . . . . . . . . O X X . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

Position 4

Click Here To Show Diagram Code: [go]$$Bc $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . X X X X . . X . . . . . . X O O . | $$ | . . X , . . X . . , . . . X . X X O . | $$ | . X X X . X . . . . . . . . . . . . . | $$ | . O X O X . . . . . . . . . . X O O . | $$ | . . O O . . . . . . . . . . . O X . . | $$ | . . . . . . . . . . . . . . . O X . . | $$ | . . . . . . . . . . . . . . . . X . . | $$ | . . X O . . . . . , . . . . W , . . . | $$ | . . O . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . O . | $$ | . . O . . . . . . . . . . . . . X O . | $$ | . . . X . . . . . . . . . . . O O X . | $$ | . . . . . . . . . . . . . . . O X X . | $$ | . . . X . . . . . O . . . . O X O O . | $$ | . . . . . O . . . . . . O . O X X X . | $$ | . . . . . . . . . . . . . O X X . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

I think the marked move above is really greedy, and is the wrong idea.

Here, I think I should consolidate the bottom and right, and just fight with one group, like this:

Click Here To Show Diagram Code: [go]$$Wc $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . X X X X . . X . . . . . . X O O . | $$ | . . X , . . X . . , . . . X . X X O . | $$ | . X X X . X . . . . . . . . . 2 . . . | $$ | . O X O X . . . . . . . . . 1 X O O . | $$ | . . O O . . . . . . . . . . . O X . . | $$ | . . . . . . . . . . . . . . . O X . . | $$ | . . . . . . . . . . . . . . . . X . . | $$ | . . X O . . . . . , . . . . . , . . . | $$ | . . O . . . . . . . . . . . . 3 . . . | $$ | . . . . . . . . . . . . . . . . . O . | $$ | . . O . . . . . . . . . . . . . X O . | $$ | . . . X . . . . . . . . . . . O O X . | $$ | . . . . . . . . . . . . . . . O X X . | $$ | . . . X . . . . . O . . . . O X O O . | $$ | . . . . . O . . . . . . O . O X X X . | $$ | . . . . . . . . . . . . . O X X . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

What do you think?

Kirby · **#994**

Not studying go this week, but an idea came to me that I want to try, once I start up, again.

Comparing study to AlphaGo, I've come to the general idea that studying high level games (e.g. pro games) is a good way to improve one's policy network - when you see several examples of "good" play, it helps in identifying candidate moves during your own games.

A big question, then, is how I can more efficiently train my "value network" (assuming the same model for learning that AlphaGo uses - maybe humans don't optimally learn the same way as AlphaGo, but I'll just ignore that for now). So what does a good value network give you? The ability to accurately assess a position and identify if it's good for black or good for white. Based on past experience, I have a very biased value network - I have some amount of arrogance, and assume that, whatever the result, it's probably good for me. Okay, maybe not all of the time (e.g. when I die), but I generally have this type of optimism.

To train this, I'm thinking of taking notes while I play. Just a simple evaluation: black is ahead; or white is ahead; or it's about even. And do this at various stages of the game. I probably shouldn't use the chat window to take notes, but maybe I can do this on the side. Then after the game, I can compare my notes to some sort of review - perhaps with the help of analysis software that gives winning percentages, like CrazyStone (or Leela does this now, too, right?).

Maybe this can help me to overcome my bias. I'll try to remember this next time I play a game. Hopefully, I read this before posting another game in this thread, and I remember to try it out.

Note-to-self: Kirby, try this out before posting another game review here.

palapiku · **#995**

It seems that humans don't have a reliable equivalent of a "value network" - direct whole-board intuition about who is ahead. Instead, even the professionals have to count each time before reaching any conclusions.

Bill Spight · **#996**

Kirby wrote:

Comparing study to AlphaGo, I've come to the general idea that studying high level games (e.g. pro games) is a good way to improve one's policy network - when you see several examples of "good" play, it helps in identifying candidate moves during your own games.

A big question, then, is how I can more efficiently train my "value network" (assuming the same model for learning that AlphaGo uses - maybe humans don't optimally learn the same way as AlphaGo, but I'll just ignore that for now). So what does a good value network give you? The ability to accurately assess a position and identify if it's good for black or good for white. Based on past experience, I have a very biased value network - I have some amount of arrogance, and assume that, whatever the result, it's probably good for me. Okay, maybe not all of the time (e.g. when I die), but I generally have this type of optimism.

To train this, I'm thinking of taking notes while I play. Just a simple evaluation: black is ahead; or white is ahead; or it's about even. And do this at various stages of the game. I probably shouldn't use the chat window to take notes, but maybe I can do this on the side. Then after the game, I can compare my notes to some sort of review - perhaps with the help of analysis software that gives winning percentages, like CrazyStone (or Leela does this now, too, right?).

Comparing notes is a good idea. However, you must remember that there is no such thing as the probability of winning the game, except 1 or 0, depending upon perfect play. All other probability estimates depend upon mistakes. So the question is, whose mistakes? My guess is that Monte Carlo estimates are no better than yours, or not much better, for players at your level. And they are probably worse than yours in the endgame. I do not know the basis of the probability estimates of AlphaGo's value network. Do they come from comparing AlphaGo vs. AlphaGo results in certain positions? Are they based upon AlphaGo's errors? In any event, it seems that some value in between the Monte Carlo estimate and the value network estimate gives the best results. Moi, I would want to know the estimates of the value network.

Bill Spight · **#997**

palapiku wrote:

It seems that humans don't have a reliable equivalent of a "value network" - direct whole-board intuition about who is ahead. Instead, even the professionals have to count each time before reaching any conclusions.

How reliable? It certainly seems that AlphaGo's is better than that of current pros. But I expect that Go Seigen's was better than that of current pros, as well.

A secret that nobody seems to talk about is that the territory estimates that pros do do not produce an estimate of who is ahead. They have the practical value of giving an estimate of how much territory you have to make elsewhere in order to win. but that is not the same thing. Players like Takemiya, who rely upon influence, have to have a good sense of its value. As an influence player myself, counting territory early in the game would just make me depressed.

dfan · **#998**

My sense is that pros sometimes make a quick assessment by counting bad moves rather than territory; e.g., "these three White stones look kind of inefficient in retrospect but everything else on the board looks pretty reasonable at a glance, so White must be behind." I'm not sure how relevant this technique is for weaker players, though; for one thing, we're even worse at evaluating moves than evaluating territory.

Kirby · **#999**

Another method I've seen is to evaluate the difference in points from a given exchange.

On an empty board, a joseki might be said to be equal for two players. So if one player gets countable profit, and the other gets influence, maybe you can say that the position is about equal. If one of the players make a questionable exchange, perhaps you can compare that to what would be more ideal and see the difference in points.

I agree that this method is somewhat mystical and perhaps prone to inaccuracies, but it's still interesting to me.

Bill Spight · **#1000**

A few comments on your last game.

Main focus: Put Black to the test.

Kirby's Study Journal

Who is online