Introduction

Consider making a contribution to Lichess.

Lichess has published over 1.6 million puzzles, helpfully categorized by themes. I’ve categorized them by the opening and ECO code that they have originated from. They are free to use.

To do this, I downloaded the Lichess Puzzles database. Each puzzle comes from an actual game played on lichess.org. The puzzles come as a CSV file.

For each puzzle, I retrieved the details of the game that the puzzle originated from. The notation contains headers that include the name of the opening and the Encyclopedia of Chess Openings (ECO) code of the game that the puzzle came from. This took at least two weeks.

The solutions to the Lichess puzzles come in a notation that computer engines use, so I had to convert them to short algebraic notation. I used the {chess} R package and created some custom functions based on it. This package was also vital for creating the board state at the beginning of each puzzle and determining which side had the move.

Finally, I created individual PGN files that contain all puzzles originating from the same ECO code. I also included the Glicko-2 rating of each puzzle, and the move number that the puzzle originates from. These files can be used with free, open source software to practice the puzzles yourself with spaced repetition.

To see how many puzzles come from each ECO code or opening, scroll to the bottom of this post.

Download

You can download the puzzle files here.

Using the puzzles

To use the puzzle files above, unzip them, and then download Lucas Chess. Other software may work as well, but Lucas Chess is what I had in mind when pursuing this project, and it’s the only software I’ve tested the puzzles on. A great feature that Lucas Chess has is the ability to use spaced repetition on any set of puzzles you want.

Once you have Lucas Chess installed and opened, click Tools -> PGN > Read PGN file. Find and select the pgn file you want to study from, and click open. You’ll see a list of puzzles, with several columns:

• N.: The nth row.
• PlyCount: the number of half-moves in the puzzle.
• Site: The URL of the lichess game that the puzzle originated from.
• Date: The date that the puzzles were downloaded.
• ToMove: Which side has the move.
• ECO: The Encyclopedia of Chess Openings code for the opening that the puzzle originated from.
• Opening: The full name of the opening the puzzle originated from.
• Rating: The Glicko-2 rating of the puzzle.
• BeginMove: The move number the puzzle begins at.
• Themes: The themes assigned to the puzzle by Lichess.
• Result: Used to indicate the end of the puzzle.
• FEN: The starting position of the puzzle.
• RowID: Row ID.

(After reading in puzzles_E60.png.)

In Lucas Chess, you can sort these columns alphabetically (A to Z or Z to A). Note that Lucas Chess does not recognize numerical values in columns like Rating or BeginMove. Because of this, a BeginMove value of 9 is considered to be alphabetically later than a BeginMove of 10.

Select the puzzles you are interested in (or just press CTRL + A to select all), and then click Train > Create tactics training.

Give the tactic training set a name (it will default to the pgn file name); the puzzle database will be created and you will get a message telling you how to find your new tactics training set. You only have to do this once.

Going forward, from the main menu you will click Train > Tactics > Learn tactics by repetition > Personal tactics, and then choose the tactics training set you’re interested in training on.

(Training tactics after creating a training set.)

Suggested usage

I included the Glicko-2 rating of each puzzle, and the move number that the puzzle originated from, because I had a particular type of usage in mind.

Some puzzles begin on the 65th move of the game the puzzle originated from. Other puzzles begin on the 10th move of the game. Let’s say you are interested in learning what type of tactical opportunities you might find yourself in playing as White as you enter the middlegame of the Caro-Kann, Panov-Botvinnik Attack. For the purposes of our example, we can arbitrarily decide to limit ourselves to puzzles that begin no later than the 20th move of the game the puzzle originated from. And let’s say that you want to begin with the easiest puzzles – a rating of 1500 or lower.

Double-click on the Opening column; you will see it change to say “Opening+”. The column is now sorted in ascending alphabetical order. Select the rows that occur before your desired opening, and click Remove. Then select the openings that occur alphabetically after your desired opening, and click Remove.

Now double-click on the BeginMove column; you will see it change to say “BeginMove+”. The column is now sorted in ascending alphabetical order (the first value is almost certainly 10 or 11). If you double-click again, it will be in descending alphabetical order. (In future versions of these puzzles, I will add a leading 0 for single-digit numbers.) Double-clicking a third time will remove any sorting logic from the column. Once sorted in your desired fashion, you can remove the puzzles that occur after move 20.

(B17 puzzles, with BeginMove sorted A-to-Z.)

This same process can be followed for the Rating column and the White column.

This is only how I’ve imagined using the puzzles; feel free to let me know how you’ve used them!

Why would I care about which opening a puzzle came from?

I don’t how useful this is; I have only used these enough to test their function in Lucas Chess.

My hope is that practicing puzzles that originated from games with the same opening will naturally put the student in thematically similar situations, and that thematic similarity is helpful for the learning process. I expect that the tactical opportunities that occur under a given opening will help reinforce what a student has learned about the strategy behind an opening.

By practicing tactics from a certain opening, the student can hopefully sharpen both their strategic and their tactical acumen.

If organizing puzzles around openings is useful, I expect that it will be because it is a complement to, not a replacement for, more established methods of learning openings.

Thanks

This project wouldn’t have been possible without the work of Thibault Duplessis, the Lichess team, and their commitment to open source products and services. Processing the Lichess puzzles was greatly aided by Caio Lente’s R package {chess}. Using the puzzles for free is only possible because of software like Lucas Monge’s Lucas Chess.

I had this idea and figured plenty of other people have must have had it before as well. Some internet searches proved that to be true, but I couldn’t find a solid collection of puzzles sorted this way. The best I found was from wtharvey.com, which was well short of the scope I had in mind, but it gave me confidence that this was a worthwhile idea.

A paper that further influenced my thinking on this idea, and inspired me to take action on it, is Training in Chess: A Scientific Approach by Fernand Gobet and Peter J. Jansen.

Other ideas

Future updates will include the over 100,000 puzzles that have been published since I first downloaded the Lichess puzzles database in May 2021. And they will probably have fixed the character encoding which is currently messed up, for names like “Maróczy”.

I would like to undertake a similar project, categorizing the Lichess puzzles around the topics in Ward Farnsworth’s (free) tactics guide Predator at the Chessboard. That is a significantly more complex task that I don’t expect to ever finish. Please steal my idea!

I would like to do some analysis on the puzzles in some of the more popular openings. For example, which squares are the most tactically active in a certain opening, and how does that compare to tactically active squares in other openings?

Replication

You can view and download my very ugly code here.

How many puzzles are there from each ECO? Each opening?

You’ll have to forgive the crappy character encoding.

For ECO:

And for openings: