Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Aquarium / MultiPV+searchmoves game analysis for ordinary players
- - By Rybka2006 (*) [us] Date 2007-11-25 23:54
Only about 10% (600 million) of 6 billion people play any chess at all, and fewer than that attempt any study. If the 10% is going to ever grow, a GUI is needed to draw more people into the game, and into enjoyment of the game.

Below master level: 99% of chess is tactics, and 99% of chess fun is tactics.

So any GUI that will accomplish growing that 10% across the world population must help ordinary people to miss fewer tactics in their own games.

Before dismissing the possibility that one chess GUI + Rybka could increase that 10% significantly, consider the Harry Potter books. Like them or hate them, so many copies have been read (repeatedly) that they have raised the literacy rate of the human race. Millions of children that were very averse to reading (and terrified of thick books) ate that series up. And as a result millions became much more literate than they would have except for one set of books.

Rybka has absolutely rocked the chess world once. If Rybka’s GUI doesn’t rock the larger world, raising enjoyment of chess and planet-wide chess literacy, then whose GUI will?

What would a GUI have to do to help a lot of people become more chess literate?

IMHO the key thing is that the GUI must make it extremely easy for the players to spot missed tactical shots in their own games.

The most direct way to do this is to show for each position in the game the best 5 or 6 moves Rybka can find. Plus, if the move played in the game is not among the best 5 or 6 moves Rybka finds, then the GUI must have Rybka analyze the move played as well.

The UCI engine-interface facilitates this sort of analysis very well.
      setoption name MultiPV value 5
will do the first part, then if the GUI sees that the move played-in-game has not yet been analyzed, then the GUI does a
      go searchmoves <the lame move played>

Let the GUI user turn MultiPV down to 3 if they have a slow machine, or turn it up to 100 if they have a monster.

Even patzers like to play some of the same lines over and over again. The GUI should support these attempts by saving a database on the user’s machine of the analysis results of previous games. If the position is already in the database of positions, just fetch the result, don’t punish the user by re-analyzing (unless the user demands it due to a new version of Rybka, wanting to increase MultiPV or deepen the ply-depth-of-analysis beyond the values used before, that sort of reason).

For that matter, the GUI should ship with that database pre-populated with a few hundred or a few thousand positions analyzed with a big MultiPV and a deep ply-depth. Understand that these are not the same as opening book positions for which there are book moves. These are the out-of-exalted-book positions that ordinary players get to over and over again. They should not have gotten to them, but they do so because most players have so little book knowledge they are out of exalted-book early.

Once the GUI has that information for each position played, that information can be presented by the GUI in a lot of ways. One set of suggestions out of infinite possibilities:

-  Don’t show notation. Show the current position in the game in a 2D board on the left, and just to the right have a vertical column of 5 or 6 smaller 2D boards. The top smaller board shows the game-move-played and its PV. The rest of the column shows the top Rybka-found moves. This allows the GUI user to use a scrubber control (like in video-editing and animation programs) to scrub the column of boards through their PV for each top move or game-played move. As they scrub to the right, every one of the 5 or 6 boards in the vertical column plays deeper into its PV, and goes backwards when they scrub to the left. Let the user use the right/left arrow keys if they prefer.

-  Then of course the user just uses the up/down arrow keys to go back or forward through the positions played in the game.

-  A blunder-check option is needed. The GUI rips through a game the user just played, showing deep analysis for anything in the real opening book, and anything in the already-analyzed-positions-database, and deep analysis for any positions in the 3, 4, 5, or 6-man tablebases (whatever the user has available). Then shallow analysis (6 or 7 plys might do) for the rest. Remember, a lot of players are out there dropping rooks and queens. A GUI is needed to quickly show them the blunderous piece drop, or the piece they failed to tactically snag.  Later, like overnight, the GUI can analyze deeper, and in the morning the user can take another look  

Keep in mind it is vitally important to show ordinary users more than the best move. Ordinary players often look at the top move and say “I don’t EVEN see why that’s good.” But when they see the second-best move, they may very well say “Oh, that one I get – I could have won a B real quick by playing that move.”

A related point is that ordinary players often have a very narrow opening repertoire. For example, they may always open 1. e4. The GUI needs to provide a way for each book position for the end user to record which moves the user approves of. We all know 1. d4 is very strong, but if an ordinary player intends to always open 1. e4 but then the player messes up and opens 1. d4, the GUI must tell the user that the user messed up.

For that user 1. d4 is a mistake, and the GUI needs to tell the user that the user played an in-book move that the user did not intend to play. Of course 1. e4 is a very simple example. A deeper example would be that the GUI user has approved that as white the user will play the Spanish, but then by mistake plays Giuoco Piano. Then the GUI sees that the user has approved for white the moves in 1. e4 e5 2. Nf3 Nc6 3. Bb5; so then when 3. Bc4 is analyzed the GUI tells the user 3. Bc4 was not approved by the user.

The GUI can use this same move-approval method to help the GUI-user all through the game. As noted above, the user might not understand the top Rybka move in a middle game, but does understand how the win-the-B move is way better than what the user played. So then the user can mark as approved that move. There are many middle and end-game positions where the top 2 or 3 best moves are very close or identical in eval score. The user can mark as approved the one that they prefer to remember.

Personally, I would dearly love a Rybka GUI that did as described above, and also let me select positions to train on with queries like
       Train me on my top 10 mistakes made in my most recent 5 games.

Of course, adding training facilities to a GUI might not be part of the Rybka-plan, but the GUI could still answer that query by dumping out a pgn file containing 10 PGN games (that’s one per training position), each with a FEN tag describing the position and the approved move. That PGN file could then be imported into a training program such as Chess Position Trainer or BookUp.

Ordinary players must improve on tactics. That is what improves the fun. Missed tactics abound in every ordinary player’s games. Someday a great GUI will find these opportunities missed by the ordinary player, and do a great job showing better tactics to the ordinary player. Ordinary players are much more motivated to look at positions they actually played, and learn that way, than to pound through a book of 1,000 positions other people played.

And by improving the tactics of as many ordinary players as the Harry Potter books have improved millions of young readers' reading skills, the GUI can raise the chess literacy and chess participation rates worldwide.

I believe that this post is both overly long and suggests a Rybka GUI that is heresy by being very different from the GUI many forum members are thinking of. But in good humor, good cheer, I choose not to apologize on either point. Instead I don my asbestos underwear in anticipation of smoking hot replies :-)

cheers, Bruce
Parent - By Zruty (*****) [ru] Date 2007-11-26 08:41
These training features were already thought of, but first we'll release the all-purpose analysis/database GUI and then will start working on chess training.
Up Topic Rybka Support & Discussion / Aquarium / MultiPV+searchmoves game analysis for ordinary players

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill