Playing against yourself

I’m thinking of buying a nice board for myself, but I know no one that plays, so I’ll mostly be using it for study.

Would there be any benefit in me playing against myself, seriously considering each move as either colour, saving the sgf then using a program like Crazystone to analyse my game to tell me which moves were good, which bad and which moves would have been better? I could then replay my game on the board while studying move suggestions by Crazystone at every step.
An argument against self play is that you can’t teach yourself what you don’t know, but by using a strong engine to analyse the game afterwards, I have a way to improve?

Don’t get me wrong, I agree the best way to learn is playing other people, which I do on OGS frequently


Of course there going to be some benefit. Amateurs are so weak, we can do literally anything go-related and it’s still going help our play.
In my case however I found self-play games boring and too lengthy.


I’ve played a handful of games that way and they just turned out to be half-pointers or otherwise really close, which was funny but it’s not terribly stimulating.

I recommend replaying pro games on a real board.


I don’t mean just playing myself and leaving it at that, I mean using an app such as AQ Go (which I’ve only just found and started using) and setting it on Leela analysis mode, then loading in a self play sgf and seeing what Leela thinks of every move I made, playing out alternate game threads if I want to based on that feedback, and so on.
I’d pretty much be teaching myself and correcting my own mistakes by pushing my self play games through complete analysis would I not?

What about self play games with Leela suggesting her best moves at every step? I could stop and think about the possible reasons for the suggestions, but I don’t have to follow them if I’m trying something of my own. That would kind of work like Pair Go with myself and Leela vs myself and Leela, with open discussion allowed!

I understand that a top bot playing strongly would suggest moves completely alien to me (24k), which is why I would still play my own games and own moves too.

1 Like

I think it’s worth a try. As S_ said, anything we do playing go will be helpful, and Leela can provide stimulus to think about moves we would not otherwise have thought about.

I’m reviewing each of my games with Leela, and I find maybe one or two really helpful ideas per game.

The rest of the variations she suggest quickly deteriorate into something I have no hope of understanding why she would do that.

Another common theme is that there’s a move she thinks is really big that I can’t see why and never would have picked, and she suggests it for every move for both players right up until the point where the real game actually took that move away :slight_smile:

But at 24k I think that a teaching game with a DDK teacher will be far more beneficial, because at TPK there are a range of basic strategies that you can be told, have explained, and follow to great advantage to propel yourself up into DDK. If you study Leela, you will beasically be trying to decipher these basics yourself from what you see - with the disadvantage that Leela doesn’t play basics.



Spot on. Totally spot on. Most importantly though, do whatever keeps Go fresh and interesting for you. I have a nice board and little opportunity to use it IRL but I’m still glad to have it and when I’m taking a correspondence game particularly seriously, I enjoy keeping track of it in ‘Slate&Clamshell.’


I would not view spending more on a nice board as an investment in a tool for improving, since anything you could do on a pricey board, you could just as easily do on a cheaper board (or perhaps just a computer screen).

However, I do not mean to discourage you from spending more on a nice board if that improves your enjoyment of the game. You should spend your money however you like.

It is also still possible to use a physical board for online games against people, specifically in the case of correspondence games, where you can just mirror the position and sit at the physical board for thinking and analysis.

1 Like

The restraint it takes not to make the title less… antagonistic… :rofl:
Nevermind me I haven’t had my coffee yet.


You all make very good points! Thank ya! I think the suggestion of using it for correspondence games as a way to sit and ponder moves is excellent, that does appeal to me. I mainly wanted to buy a nice pricey board just for the joy of owning one.
I could buy a foldable plastic magnetic version for £12 but just wouldn’t be the same!


I have no inhibitions, at the moment, so I may have to make a comment…

In the words of many a parent (from what I understand, anyways): Quit playing with yourself. You’ll go blind.

That said, I think that statement can hold strong for Go. I feel that you don’t make an honest guess at what your opponent would play, only what YOU would do in a given situation.

However , I feel that - as has been mentioned previously (and in my lowly DDK viewpoint) - pitting said “opponent’s” moves against a good AI could help. I’ve been told by a few dan level players on twitch (not directly, but just has been said) that the computers play weirdly (compared to humans), and that may affect your game.

Sorry if I am rambling…I don’t always think things through (look at some of my past games…just saying)

It depends on the AI a little. Some of the weaker monte-carlo AI do tend to make strange moves (by which I mean quite bad moves) from time to time. If you get used to those strange moves, it gets easy to beat them even though on average they are ought to be above your skill level. For example, there are certain joseki that are advantageous to you because the computer will consistently misread a ladder, so you can give yourself a reasonable head start by starting like that.

As for the strong AI’s (which is probably what your twitch players talk about); they don’t necessarily play weirdly, but they play many moves that don’t align with basic go strategy. As a beginner it is more helpful to get a good feel for strategy than it is to dive in the complex moves that AI make.

A computer doesn’t know the difference between playing basic safe moves and playing advantageous dangerous moves. So if a computer has the choice between enclosing his corner or do a very early (seemingly reckless) invasion, it might very well turn out that the early invasion appears better for the computer. Not because it’s easier to win, but because the computer has read out the several thousand of ways the fight can go, and discovered that overall it gives a better advantage than enclosing the corner.

Meanwhile when the beginner mimics this tactic, he easily screws up the invasion because it gets too complex, makes a move in gote and has his opponent not only destroy the invasion, but also approach the non-enclosed corner.

Another problem with reviewing with an AI is that they don’t explain moves. Do they tenuki from a situation because they think it’s hopeless to play more, or do the tenuki because they think they are solidly alive regardless of your move? Why did it protect against this cut over here, but left the cut over there open? Why did it pincer instead of back off in this fuseki? The answer to these kind of questions is probably what’s the most useful for a human beginner to understand. At least more useful than knowing what the best move on a board is.

That said, I did play against bots for about 1.5 years before I started playing online, and it brought me (together with watching the NHK cup, dwyrin’s series and getting lost in sensei’s library) to around 10k.


I play against myself fairly often, and while this has obvious limitations I find it less predictable than one might suppose. I have little doubt that I’ve subconsciously steered some games in a particular direction to justify one harebrained scheme or another. Yet over time this has instilled in me a certain waryness, so that nowadays I’m more likely to test my hypotheses in the most severe and inconvenient way I can imagine. And I like to think this has benefited my game somewhat.

Now maybe I’m just kidding myself with this. What I can say for certain is that I still find the games compelling and enjoyable, a worthwhile exercise. The games do run a bit long since they seldom end by way of resignation. My opponent and I know each other a little too well for that.

I haven’t gone blind yet, though I must admit to bouts of tunnel vision.