Database Review
A big picture approach to studying chess
In chess we’re used to computers being the authorities on best play. It’s been 26 years, after all, since Deep Blue beat Garry Kasparov. Yet in poker, it’s only in the last few years that solvers (the poker equivalent of chess engines) have gained widespread adoption. They’ve brought with them a revolution in strategy.
But by my reckoning this is the second time computers have revolutionized poker. The first time was when online poker made it feasible to track large numbers of hands with perfect precision. Before that, the only ways to work on strategy were logic, intuition, and your (highly imperfect) recollection of what happened in the hands you could remember.
It turns out that poker is a very hard game to understand one hand at a time. The luck factor is huge in the short term, so the results of any given hand can easily be misleading. Only over a large number of hands can you see the true contours of winning strategy emerge. Between the forest and the trees, poker is about the forest.
In contrast, chess is a game of specifics. With no hidden information or randomizing element, it is sometimes possible to calculate what will happen in the future very precisely. Take this position from one of Eugene’s games at the Massachusetts Open. Can Black take the pawn on d4?
No, seriously, can they? See if you can figure it out.
In fact, Black can get away with taking the pawn, but it relies on a series of clever zwischenzugs (I think this might be the first time I’ve ever referenced an actual zwischenzug in this newsletter!).
27... Nxd4+! 28. Ke3 Nb3 First buying time by attacking the rook… 29. Rab1 d4+ Another zwischenzug to vacate the d5 square. 30. Ke2 Bd5 And all of a sudden everything is defended. With four pawns for the exchange Black is winning comfortably.
The game often comes down to a detail like this.
In poker, it’s generally not possible to calculate the possibilities with this degree of precision. There are just too many possibilities. For example, if you find yourself on the flop in No Limit Hold’em, there are 45 possible turn cards, followed by 44 possible rivers. That multiplies out to 1980 ways the remaining two cards can come out. In terms of raw numbers that’s not so different from chess, where there could be 30-40 moves in a middlegame position. But in chess, most of the legal moves can be ruled out. In poker, all the cards are equally likely to be dealt. You can’t calculate all of those possibilities.
What’s more important is to make sure the overall shape of your strategy makes sense. Are you taking each action with a mix of hands that makes it hard for your opponent to react correctly?
That mix is key. The poker version of a chess engine is called a solver. When you look at solver outputs, they often contain mixed strategies: hands that are played in different ways at some frequency. For example, the solver might want to check a certain hand 80% of the time, but bet 20% of the time. This means that if you’re reviewing one hand at a time, neither checking nor betting is a mistake, but they only form an optimal strategy when mixed at that frequency. The viability of your strategy is simply not visible one hand at a time.
Do you have to copy the solver strategy exactly? Not necessarily, but if there’s a spot where you’re bluffing too much (or not enough) and your opponents know it, that can be a big problem. In this scenario, you have to take the global view.
The big picture view isn’t as evidently necessary in chess as in poker, but it’s still extremely valuable. Isolated mistakes can be just that – isolated. But some clusters of mistakes reveal thought process errors that, if corrected, can level up your game. A knack for sussing out these systematic mistakes seems to be a common factor in players who are able to improve rapidly. Such patterns can’t be seen one game at a time; they must be connected over multiple games.
If you find the whole idea of thought process a bit too abstract, there are also more direct applications of big picture analysis. How well do you score with each of your openings? Are you typically ahead or behind on the clock? Do you manage to save losing positions as often as other players at your rating level? These sorts of questions are very hard to answer ad hoc, one game at a time. Biases will almost certainly distort your perception of what’s going on. But if you play online (or keep a database of your over-the-board games) you have a record of exactly what really happened. From there it’s easy to calculate the answers to these questions.
I had initially intended to try one database analysis as a trial run, but I got so many responses that I decided to commit to four database reviews. To start with though, I’m kicking it off with one person chosen randomly from everyone who responded. Check out the results in the video below:
I already replied on youtube how I liked the part on time management, because I now realize how much sense it makes in my games too. I am blundering and making less inaccuracies because I am taking my time more in hard positions. I am pretty sure I would be a better player if I just learn to convert in a more efficient way than trying to rush the middlegame. However, I had never tried to do as Nate suggest in the video and look at my times spent per move. I found in my last few games I was spending way too much times on certain bishop moves (to e2 or d3 ?) so it is definitely something I will add to my game analysis now. Thank you Nate !
The video is very interesting. I agree that the automated graphs/stats that these sites give are not that useful to most people. The way you looked at time management makes sense to me intuitively. I wonder if the fact that you are looking at averages over many games might make the graph skewed somehow. Might be you want to visualize how far from the “perfect” time graph each game is (or calculate a number for how much they deviate from the optimal time usage. ). Kind of like the accuracy score but only for time usage.