New Tactics Features: Stream of Consciousness, Train your Weaknesses

Introduction

I wanted to run a pair of ideas by the forum to see if they are actually worth pursuing, and perhaps brainstorm further and build upon it together. The ideas are fairly simple, and I feel they address a frustration with the link between tactics solving, pattern recognition, and long-term retention.

Before going into the ideas, I would like to share a story. In the past, I have used a language learning online platform called memrise, and a feature that stood out to me in the learning process is that your weaknesses are emphasized and reemphasized by frequently repeating the content within short bursts and in different formats (in the case of language learning, oral, written, multiple choice, etc.) and the combination of these different formats and repetition reinforce the linguistic patterns in an incredibly short timeframe, specifically because the more frequent repetition of the words you failed at closes the gap between what you've learned so far, and what you have retained. I was able to achieve an A1 level in German with 2 months of training 2 hours a day. Years later, I still remember most of those words I learned, and the different contexts they fall into, and I feel like this is something that can be achieved in chess.

I would like to note that my ideas are different from memrise's, however they uses the same fundamental principle. Tactics fall into categories, just like words, and those categories are expressed in words (Mate in 4, Zugzwang, Perpetual Check, etc). Some are more fundamental, and some are more complex, but a branch can grow from the simple to the complex with time and repetition. The current tactics trainer allows you to train specific ideas unrated, and that's fine, however, it only allows you to learn in one way. So what are the alternative ways of learning tactics?

-

Idea 01: Stream of Consciousness

Watching people play chess is great. A lot of people on the site seem to do it and enjoy the duality of chat and human battle. But when it comes to tactics, watching high rated players solve on puzzle rush on twitch once in a while isn't really enough to understand the complex ideas of the higher rated puzzles. There are also simply too many different kinds of puzzles with different kinds of ideas behind them, many of which could be irrelevant to one another across sets of 10-20-40-etc. A healthy feature on chess.com tactics is the tags. While we're on the subject, it's a shame that I can no longer see the tags when I fail a tactic, because those tags were a helpful way for me to associate the puzzle with the ideas behind it. Using the tag feature, tactics can be split into these categories and viewed separately in... video format.

Rather than solving the puzzles yourself, what is essentially a chess board slide-show with ideas being highlighted, then played out, at a speed of your own choosing (.25x, .5x, 1x, 2x, etc.) will guide you through a series of tactics with the same theme in them. The UI idea is a set of icons representing each of the different tactics themes that a user can press on and choose a rating range and whether they would like the tactics to flow incrementally upwards, downwards, or randomly within the range set. The "video" chessboard loads, and begins to play out tactics automatically.

The position plays out with the new analysis symbols representing whether the moves were mistakes, brilliant, etc. The method is as follows:

00. At the start of the tactic, we see all information about it (rating, avg. time, etc.)

01. Opponent makes move. Move is highlighted as a blunder.

02. Weakness square is highlighted on the board.

03. Move arrow is drawn out from the piece to the weakness square.

04. Arrow disappears, move is played.

05. Repeat 01-04 until puzzle is done.

06. Original position reappears.

07. All moves that were played are highlighted with overlapping arrows to illustrate the patterns.

08. Fade to next puzzle.

This would be an excellent way to train pattern recognition in relation to different ideas in chess tactics. The idea uses the Leitner flash-card system to reinforce learning and chess visualization in a calm and relaxing way. The brilliant thing is that you can watch tactics the same way you would watch any other video, and you can set the parameters and even have parameters suggested and catered to you (similar to work-out apps curating workouts based on progress). The potential is endless for learning, and the UI is fairly simple to execute, since it simply pulls out data from the server, and plays it back in an automated fashion using features that already exist.

-

Idea 02: Train Your Weaknesses

As previously mentioned, this feature on memrise helped me overcome irrational but fundamentally personal difficulties with understanding certain words. Chess.com already has the data on how we perform on tactics trainer. Many problems tend to repeat, and sometimes you pass them, and other times you fail them. The way to deal with this is simply to more frequently repeat that tactic until it becomes second nature. However, the way rated tactics works doesn't allow this, and since the body of information is huge, this makes it difficult to see the same tactic again anytime soon. In fact, sometimes I won't see a tactic I failed again for hundreds of future tries, and this is somewhat unfortunate.

The UI idea for this involves having a separate button in Puzzles, "Train your Weaknesses". Chess.com automatically pulls out tactics you've failed previously, and presents the set to you repeatedly until you begin to solve those puzzles correctly. In the future, the puzzles you get right more often will begin to appear less frequently in your training globally. The UI could be similar to the current rated tactics training. The difference between the two training methods would be the following:

Tactics solved in "Train your Weaknesses" mode will still give you + or - rating, but half what is normally awarded. Since the probability of a puzzle repeating is higher when training your weaknesses, halving the awarded or reduced points will ultimately level out your rating once you begin to solve the puzzles correctly. This helps to build neural connections that last in relation to specific ideas you are having trouble with.

-

Well, that's all I have for today. I hope you guys think it's worth pursuing, and I'm looking forward to a dialog on the subject!

RolloOrollo

Chris

You should play more Puzzle Rush. That shows you the same types of problems over and over.  I am approaching 850 tries. You are not quite at 400.

I honestly feel like I am seeing many combinations more quickly during games.

Puzzle Rush is great, but it's a whole other feature with its own method of learning. There are other methods too, and I think we should be open to experimenting with them, don't you? @notmtwain

Thanks for suggestions! Will have a look.

Thanks for the consideration @jdcannon

I like both proposed ideas. Hopefully someday chess.com will be able to implement them. 
Probably the main challenge here is to create robust classification of patterns which would be easy to remember for many people.
For example, "Mate in 2" pattern is not that good because it includes lots of completely different ideas.
Form the other side, using specific combination of few attacking pieces seems much easier to remember.
Puzzles [184637, 502400, 431543, 644920, 141679, 350445, 619074, 186397, 38064, 560076]
can be included in the same pattern which can be named something like: "Pawn attacks".
Puzzles [609584, 118208, 589192, 497602, 561190, 132855, 346205, 563544] can be named as "Queen Knight Attack"
Puzzles [84019, 496222, 479862, 745114] can be named as "Rook Bishop Knight Attack".
In addition to the patterns related to various combinations of attacking pieces some other trapping related patterns seems easy to memorize also.
Like: "Trapped Queen" [ 509266, 35642, 35866, 45425, 30461, 361381], "Trapped Rook", "Trapped Bishop".
Another easily memorizable pattern may be related to non-trivial "Back rank" ideas [531720, 697588, 75338], various "Pins",
"Discover attack", "Queen|Rook|Knight|Bishop Forks", "Overloading", "Remove defender".
Sacrifice patters like "Rook Sacrifice" [ 648304, 521276, 41664, 562922, 501572], seems pretty useful also.
Even things like "Unexpected Check" [ 89033] could be combined in a separate class of patterns,
based on its statistics, only 62% people solved it.
"Zugzwang" is kind of important pattern of course [640872], but it sounds too general to me, like "Mate in 3".
Though I have no idea how to break it down into easily memorizable patterns.
The bottom line here is that the success|failure of such pattern recognition tool will be mainly
defined by robust|poor classification of implemented patterns.

That's a great set of ideas @OldHatGrandpa

I can already see the Puzzles I Have Failed section in training which is very useful (was it always there?). It would definitely be useful to analyze the positions manually and classify them as you said with metadata. Chess.com already does what you are suggesting to a certain extent! When several tags are present, a tactic will have all the data related to it attached, so you can see it's a:

Mate in 2

Discovered Attack

Pin

etc. but only in training mode as of the UI update.