The idea of deliberate practice
Deliberate practice consists of studying some aspect of Go and then implement it in one's next games, to review and improve on that very aspect, then move on to other aspects, while maintaining the strength obtained in previous cycles.
Many players seem to think this is only possible through recreating the same board position, hence only by replaying the same opening. I think deliberate practice is not so much about specific positions but about concepts, attitudes and boundary conditions.
The best way to improve is having a teacher. Teachers will select the next subject to work on and monitor the implementation. Teachers provide an external source of motivation and positive or negative feedback.
It is harder to reinforce a good implementation through positive feedback in absence of a teacher, because game results are the compound of numerous decisions, not only the one in focus. This is why I wanted to assemble a learning track, in which the aspects taken in focus, are progressively making the biggest difference for the game result, not in every single game perhaps, but over a streak of games.
In the classic approach, training and competition are only linked through the subconcious, by drilling good practices which next become a natural part of the game. I believe this is true for tsumego and pro game study. The idea of deliberate practice I explore, goes beyond that and involves the conscious into the feedback loop.
What really makes a difference
I've written about this in the /PlayingUpToYourLevelOfUnderstanding, where I've discerned the following aspects as making the biggest difference in any players' games:
- don't resign
- don't lose on time
- don't play games in dire conditions, be it poor internet connection, being deprived of sleep, having drunk alcohol or otherwise not being in an acceptable state
While the first one may go against sportsmanship, it is important to develop some fighting spirit. There will be time to become a gentleman or lady later. The second one, on top of reducing stupid losses, will save players from playing random blitz games which don't allow for any kind of deliberate practice in the first place. The third one will increase the level of concentration.
Bottom line, these first three kinds of deliberate practice focus on reducing stupid losses. It has occurred to me that many players, including myself, lament about stupid losses in an otherwise advantageous game. As a differentiator, stupid losses are much more significant than "poor implementations of the micro Chinese". So, any amateur player who wants to generate positive feedback from games, should start out with reducing stupid losses, not with practicing the micro Chinese.
I realize that the conventional interpretation of "deliberate practice" is to consistently exercise at a higher level than your current one, to stretch your abilities. I found this approach not very productive in Go, with all its pitfalls and opportunities to blunder. So I want to get things straight at the basic level, before moving to "real" study and practice.
Put it into practice
It's all very well to make such statements, but are they really true? I thought I'd put myself to the test and play 10 games with the above threesome as focal points. Don't resign, don't lose on time, play in good conditions.
- In my first game, I played a 1k, with 35 minutes sudden death. I lost on time! Lessons learnt:
- Avoid sudden death games because they are more blitz than they appear.
- In sudden death games, the determining factor is whether you have more time left than your opponent. They are really all about time, not about Go.
- Even so, I was behind on points. Positional judgment and basic technique can improve.
- Second game: 10+25/4 against a bot 1k, +R. Lessons learnt.
- Played too fast in byo-yomi this time. Took many risky decisions.
- Basic life and death missed early, then found later.
- Bot couldn't play ko.
- Third game, against 1d in 10+25/3. Won +50.
- Good time management.
- Two awkward noseki variations. The small nadare one went into a favorable ko for me.
- Most importantly the guy kept on playing with 50 points behind, but it didn't bother me because it's the first amendment: don't resign. So, when you have decided you won't resign, you're less troubled by an opponent who doesn't. It's an exercise in stamina.
- Fourth game, 45+5x1 against GnuGo 3.9.1 (no rank). Useless game at technical level but interesting for the current focus: the bot played way too fast for the set time and I followed along. Lesson learnt: don't lose on time but don't throw away your time either.
- Fifth game, another bot. Stupid liberty counting mistake deep in the game, lost by 2,5 points. Lesson learnt: count liberties and remove aji to win. lesson 2: don't play with headphones.
- Another one against the same bot. No music this time. Using all my time. 30+
- game 7: lost a big group and resigned when there was really nothing to do. Lesson: look at connections.
- game 8: play for connections and thickness. Easy win.
- game 9: play for connections and thickness. Easy win.
- game 10: crushing a bot. This gives me a 7-3 score. Losses on time, resignation and by 2,5.
Lessons learnt (should become natural in next series):
- Basic gamesmanship: play decent time settings and use your time. Don't resign. Don't lose on time. Avoid distraction.
- Basic strategy: connect and cut.
- Basic tactics: count liberties, especially late in the game.
- Basic technique: basic instinct under time pressure.
- Major lesson: take time to think when the going gets tough.
- Did all the easy problems at gogameguru. There are a few tricky ones in it! A good exercise to start off with 120 problems that should all be solvable in a few seconds. They contain most of the basic tesuji for life and death and connection. No dan player, ex- or aspiring, should feel too good to rehearse them. Next I'll do the intermediates.
- Trying the intermediate problems at GGG. I'm missing a lot of them and resolve to trial and error. I will have to take a step back and think longer about them. They are really far from easy.
- So I resolved to Beginner exercises here, first 50. I think everybody should do these once in a while. They're very valuable material. Incidentally, I missed beginner exercise 18.
- Beginner exercises 51-100. 55 and 95 tricked me.
- Always parallel star point opening.
- When defending against the approach, make a one space jump.
- When pincering, use the one space low pincer.
- When a group is surrounded, carefully study its life and death status.
- Kill when possible. Live when needed.
- Tenuki when neither.
- Consider basic haengma:
- diagonal for steady development in dire conditions
- one space jump for basic defensive development
- knight's move for attacking development
- Lost by 3,5.
- Don't play out ladders that don't work, even if it looks like a fancy squeeze
- Look for miai in the cuts, instead of crudely cutting and lose the capturing race.
- Lost by 20+.
- A double atari may look like a ko but it really isn't. Stop making beginner mistakes.
- Don't try to come back in a rush
- Won against 1d bot +R.
- Won +R
- It can really be easy if you just play normal moves.
- Enter frameworks from the border, don't try to attack their walls at vital points.
- Won +R against 2k: good squeeze technique, usage of aji
- Won +R against 1k: superior L&D skills. Initially was behind due to wrong joseki (one of those few where a mistake can make a big result). Later he killed an invasion but I was able to use its aji to the fullest.
- Won +R against 1k: again superior L&D skills, some sacrifice, hence good ko fight. It's time to increase the level, these games are being too easy. It may be that the tsumego is paying off. In any case, online it does pay off to surround and avoid being surrounded.
- Won +R again against 1k bot. Again ko with superior L&D skills.
- Won a blitz game +R against 1d. Didn't realize it was that fast. Big moyo. Kill invasion.
- Came from behind in a sanrensei moyo game, where the opponent converted thickness and sente well into side territory. My corner invasions were not enough. Fighting in the middle went wrong. Due to a few kos I came back and my opponent resigned when the score was actually even! Like I say: don't resign if you don't want to lose. Anyway ... that completes 10 more games.
- With a score of 8-2, after 7-3 in the first series, I win 3/4 of my games, so I'm probably a bit underranked, which would make me 2d (again). Competition can be tougher.
- Nevertheless, the winning ratio is due to two reasons (I feel)
- I play with good spirits, don't expect to win but never give up. I hang in there and have reversed at least 3 games.
- I have a better L&D judgment than my opponents. Amusingly this especially gives me an edge in ko fights. I always emerge victorious after a ko fight (last game, the opponent's threat was no threat - it took just a small bit of reading to find this out).
- I don't have issues with time usage anymore. Don't play too fast, too slow, I'm using my time. In blitzish games I tend to not use all time available, but then I remind myself to do so.
- Overall, my impression is that deliberate practice, with a focus on what to do to win games is paying off.
- In the meantime, doing lots of easy L&D problems is also paying off, be it somewhat more unconsciously. The increased confidence in reading basic positions is palpable.
- Beginner exercises 101-200. A few tricks again. In general the beginner exercises are 95% beginner, just a few of them are too difficult for beginners. Quite a few are contrived. The gogameguru series is better structured, but overall SL hasn't done a bad job at all.
- Beginner exercises 201-250: more contrived ones, difficulty level going up but still easy for me.
- Went through beginner exercises 251-300. This time around, I needed to correct a couple of them, while I also felt the difficulty level goes well beyond beginner, especially when Reuven's tsumego from games kick in. Maybe it's not that bad though, because after 250 beginner problems, I think you may be ready for some more challenging ones, so why not. At least for me the exercises are a good training of basic shapes, ko techniques and overall the methodical thought process of tackling a position, assuming there might be more to it than meets the eye.
- Done all 340. The last ones added are nice ones!
- Time to tackle gogameguru's intermediate ones. (target for this series)
- I also intend to tackle the elementary series in the Hitachi go problems. They're at 953 now, so that should give a good deal of work :). It seems that their difficulty is beyond our beginner exercises but from the first ones I've done, it's definitely solvable for me within a minute. (stretch)
- Based on Antti's thesis, I might do some rote learning of games. Perhaps I'll go back to the games of Otake Hideo I once studied, but I might also pick the current Lee Sedol - Gu Li jubango. (stretch)
- Another ten games, bearing in mind all the above lessons learnt and trying a few new things. In the games where I got into trouble, I had floating groups in the centre which my opponent attacked for good profit. I might postpone 3-3 invasions for some more time and focus on thick groups, maybe go seigen groups, and return to one of my original objectives, to get the game into a comfortable endgame. Therefore, I might
- Focus on thick groups
- Keep track of the score (primary goal)
- Keep it within a 10 point (dis)advantage
- Study endgame techniques again (that will be for series 4)
- Unimportant game. Victory. Opponent was too weak and the kill was huge and easy.
- 1k opponent with a very tough fighting style. I was behind but hung in. In the end I pulled of a tesuji and he resigned. When I looked at the position, it was 0,5 point.
- A similar thing happened with a 1d bot. I was behind in the middle game but came back and then the bot resigned while the game was not over. Shouldn't play bots perhaps, even if they're ranked 1d. The lessons learnt remain positive about the stamina, but on the negative side, I tend to get involved in big fights which are to my disadvantage. I'm playing bamboos which have little eye potential. Something to review with a stronger player.
- +R against 1d. Lost a big group in middle game due to bad L&D reading. Revived it with very good shape insight (and some mistakes by the opponent) and captured the surrounding stones. Never give up!!! Back to 2d.
- +R against 1d - easy
- +R against 1d - he made a basic ladder mistake after I got severely behind in early fighting. Really, staying focused until the end is the key to victory.
- Live game with 5x30s, -3H against a 5k from the Ghent club. A double kakari variation turned into a big dragon that could either be cut or turned into a L group. Still, the large capture had to be maintained.
- Same settings, same opponent. After a lot of fighting and mistakes at both sides, the game ended in 6 points in my favour.
- +R against 1k bot, 2H Blitz game
- +R in a free teaching kind of game. The game was unbalanced.
I included the last game in the series because I want to start a fourth series with the same objective as the third, because I feel I didn't really practiced what I wanted. I was merely surfing on a victory wave against players who didn't really put up much of a fight (including 1d bots). So I'm going to add another objective, which is playing (human) 2d opponents only.
- Done gogameguru intermediates 1-60. A few I could not find, like 49. Sometimes when I find it, I click for confirmation and find out about variants or about a one way street that comes after a correct first couple of moves.
- Done gokyo shumyo 1-20. The first few are pretty basic but the level rapidly increases.
- Done gogameguru intermediates 61-80. These become rather difficult. Very good to revisit. 1 minute to 1,5 minutes per problem.
- Done gokyo shumyo 21-50. Less than a minute per problem. About 85% correct.
- Done gogameguru intermediates 81-120. Easier this time, also due to the effect that I remembered the more recent ones. Still, it's surprising to see how certain tesuji tend to stick even after a long time.
- Did about 50 problems in the http://www.101weiqi.com/10K/ series. Even if 10k cannot be expected to put up a high barrier, I wanted to work my way up from the basics again. However, I might drop this source again. The quality of the problems has too big a variance. The same is true for the level. Some 10k problems are about basic nakade, others about how to squeeze more points out of a situation. Recommended by Hushfield? though, so I might try one of the higher categories.
- Switched to the Korean problem academy at gobase, thanks to a tip from Bram Vandenbon. Did 1-30 of the 5k-1k series. See http://gobase.org/studying/problems/academy/.
- Game play
- Play human 2d opponents only
- Focus on thick groups
- Keep track of the score (primary goal)
- Keep it within a 10 point (dis)advantage
- Done first 15x2 Hitachi go problems. Great series. Some from gogameguru come back here (or rather the other way round)
- Quick review of endgame techniques here.
- Done Hitachi go problems 16-30 both Elementary & Intermediate. Some techniques like oki wants ikken tobi are becoming intuitive. Wonderful how 30I depends on 29E.
- Done Hitachi go problems sets 31-64. The ko problems in the corner are difficult for me.
- Done Hitachi go problems sets 65-112. These series are intricately building on each other, so that I really get the impression of becoming smarter. An example is problem 108, one that I wouldn't have been able to solve from scratch in days but became suddenly solvable after doing problem 106. Many of the "intermediate" problems are still too hard. I think I solve about 60%. This doesn't frustrate me at all because I can do them again while they already provide me with good ammunition.
- +R against 2d. I played Black with a fairly dull ni ren sei opening. I kept my groups thick and kept track of the score. My opponent did not yield in the endgame but resigned when 5-10 points behind. I didn't expect it but I knew I had this much of advantage. I saw a few opportunities he missed but don't know where to play better myself. This looks pretty much like my level. In any case, all deliberate practice well achieved.
- +R H2 1k bot
- -11,5 against 2d. 5x20 game. Slightly too fast perhaps. Got behind by playing a joseki wrong but in the end lost because I failed to see a ko option on a big corner.
- It had been a while since last game and I had an opportunity, but no 2d was available. I accepted a 1d challenge but the game was too easy. +R again. There seems to be a big gap.
- -9,5 H3 2k bot
- Fighting game against 1k. I lost -R with 2H.
- Easy game against 1k bot. This series is not going very well in terms of finding real 2d opponents. 3 more games. Let's try to play decent games here.
- -7,5 with 1H against 1d. Playing too passive under time pressure.
- -R against 1k bot with 2H
- +R against 1k bot with 1H
After 10 games I don't seem to be able again to reach my objective of playing 2d humans only, keep track of the score and play thickly. I even regressed into not dealing very well with the time conditions.
Neither did I spend a lot of time on tsumego lately. So the level is stalling, at best. A lot has to do with more work at work and a lot of focus on table tennis. There are only so many things you can do at once ...
Back to 1d and not so able anymore to win easily.
- Don't resign
- Play games with enough time
- Use your time to the full period but don't lose any period in the opening
- Consider 3 alternatives for each move, as a habit
- Calculate 3 moves deep (at least) for each alternative, as a habit
- Count liberties of all groups involved, especially towards the end of the game
- Give priority to strengthening groups
- Pause when a group is surrounded, to assess its life & death status. Give away a full time period for each group.
- Pause at the start of the endgame to identify. Give away a full time period.
- Try to get the majority of the big endgame points
- 1d win
- 1k win
- 1k win
- 1d loss
- 1k loss
- 1k win
- 1d win
- 1k win
- 1d win
- 1d win
- 1k loss
- 1k win
Analysis of major mistakes
- 120 - timid move that allowed opponent to enter major territory; root cause: not calculating the score & fear
- 52 - aggressive move that allowed opponent to cut me on big scale; root cause: impatience to capture; not strenghtening big group first
- 84 - wrong haengma for connecting; root cause: reading
- 56: timid move that allowed opponent to break up moyo; root cause: failure to see own strength
- 128: better sacrifice sequence possible; root cause: take small stones and not big scale
- 12: wrong haengma to capture pincer
- 70: accept good shape; root cause: originality
- 90: not cutting off major black group; root cause: not pausing to check global position after local success
- 66: entering enemy territory with weak group; root cause: greed
- 46 attack cutting stones from the wrong side; root cause: failure to see which group is strong
- 25: not connecting; root cause: make territory while attacking from the strong side !!!
Major lesson after session 5: understand which of my groups are strong and play accordingly.
Purpose: exercise /time management.
- Siberia 1d +1,5
- Bam 1k +R
- Alouis 1k +10,5 Good time management, using the last period for endgame execution, making some good decisions in middle game by using the rest.
- Aemv 1k +15,5, promoted to 2d again. I had one unused period. Played a little too fast perhaps but the game went well.
- Zion 1d -R. The settings were 5x20 and that's not enough for me. I went into last overtime period too fast. Anyway the game was lost due to a reading contest where I was outsmarted.
Session 7 (June 2015)
I've been playing a lot of games lately, very often against robots. The score in May and June was 20-18 and as such I did not improve my rank (quite the opposite, it has dropped a bit). More importantly, I don't feel like I'm improving either.
Bots like Hirabot are very strong in fighting but in some cases are outsmarted in complicated life and death. I find it more challenging nowadays to play them.
On the other hand it is clear that simply playing isn't doing anything. So I've started training again.
- 133 easy problems at gogameguru. Spent about 100 minutes solving and scrolling.
- Play every game from now on with the time management practice in mind. For the sake of repetition:
- consume 1 period for a major tactical or strategical decision in the opening or the early middle game
- consume 2 periods for major life and death calculations or a ko fight
- consume 1 period to calculate the score at the start of the end game and to identify the major endgame moves
- almost use every other period, say at least 25 seconds, to calculate as many local possibilities as possible or to increase global awareness
- 1m + 5*30s is an ideal time setting
- after every game write down:
- did I manage my time according to plan?
- how did I fare with the major decisions for which I consumed a period
- did I make a good positional judgment and overview of endgame moves
- what was the winning/losing move?
Session 8 (July 2015)
Same plan as session 6. The results are mixed. Most of the time I'm challenging bots.
In the meantime I'm redoing the intermediate problems at gogameguru.
Session 9 (October November 2015)
Good results with playing thickly and winning the endgame. Learning a few things about the endgame too:
- endgame which affects life and death status of groups is always bigger than it seems
- some territorial sente are incidental to the endgame
- evaluate what you think are the 5 biggest endgame and re-evaluate after 2 of them have been played out
- it doesn't really matter how big a move precisely is if you get the general flow of moves right
- continue to reassess life and death of groups
I also replayed a couple of Gu Li - Lee Sedol games to assess my endgame evaluation against theirs. This was a very instructive exercise as I learnt more about the aji left in their "stable" groups and the relationship between endgame moves.
Results against human opponents:
- 4 wins by resignation
- 1 win on points (24,5)
- 2 losses by resignation
so no real tight endgames yet against humans.
What science says