YOUR Use of Engines for OTB Play

[leavenfish]

I want to start this thread to have people give tips to one another on how they use engines for OVER THE BOARD (OTB) preparation.

I use engines to shore up two types of OTB play: Longer time controls (like Game in 2 hrs) and very short ones (Say Game/45 min...very popular locally).

Ideally I like to run an engine over my repertoires in MULTI-PV mode, looking to steer towards lines where my opponents have MORE (sometimes just subtle) ways to go wrong. OTB in particular being a game of mistakes...so give your opponent more opportunities over the course of a game (specifically the opening phase) to make incorrect decisions!

Doing this manually - often in Chessbase and using the 'pass function' (x) to see what kinds of threats I might be missing in a given position is helpful.

Generally...I just tab backwards or forwards thru my openings in multi-pv mode, looking for situations where my opponent (mainly) has a fine line to walk - where maybe only 1 or 2 moves are 'good'. Then I note those which may look reasonable, but are not...and do analysis on those to find out to best exploit those moves.

I then throw she Opening Repertoire into something like Chess Position Trainer (Similar to COW or Bookup...just looks a lot nicer and fewer bugs) and then train myself on memorizing the lines I have worked out. I often tab back and forth within CPT doing the same.

It is all quite tedious...but kind of enjoyable. Backsolving options (like exist in the programs I mentioned) can be of use, but really only once I have put together a pretty good opening to begin with.

I very well may be missing much and would love to see how others use engines to help in their OTB opening preparation. You thoughts?

[leavenfish]

So, people on here just like to play engine matches? Not use Komodo or Stockfish or Houdini ...or any other engine in any meaningful way to further their opening preparation for OTB play?

[MikeGL]

I very well may be missing much and would love to see how others use engines to help in their OTB opening preparation. You thoughts?

In my opinion, only top 200 GM's can make use of deeply prepared opening lines. To improve my OTB play, I sometimes play against engines
on endgames K+Q vs. K+R, play difficult random middlegames, etc.
But mostly it is endgame training, like R vs B endgame or B vs. N endgame.

Studying a chess opening is just memorization, rote. No critical nor analytical
thinking involved. You'll be stuck in your rating and won't improve much
if you concentrate on opening studies. Just research and learn how all
the Polgar sisters became GMs at a young age, and you will get my drift.
Hint: they didn't play hundreds of thousands of online bullet games or blitz games, nor they deeply studied opening lines.

[matejst]

I planned to answer to your question, since I use only engines as a way to prepare for OTB play, but I couldn't find the thread any more.

1. There's a part of the preparation where you don't use engines: solving problems, reading games without a table, blindfold games help you to improve tactically.

After that, you can play part of games or complete games against tactical engines, like OpenTal or Pro Deo Q3. Losing is not a problem here: you will learn to fight negative emotions, and stay focused even after a blunder or a bad move. Analyzing these games, after that, with an engine is highly recommended.

2. A good repertoire is a condition sine qua non to obtain good results OTB. But instead of focusing on lines, better try to focus on ideas in a repertoire, and it's an imperative to do the analysis yourself. I use older, slower engines when I analyze games for my repertoire, because I can understand their evaluation process and I can myself calculate with them. In a tactical position, I will probably use Gandalf: I'll do my own analysis, switch the engine on, and check its answers... and mine. At a level below 2200 Elo, a double edged position, even inferior, is not something to avoid if you know the key moves/ideas in attack/defense.

3. Engine choice: modern engines are mostly useless: they are too fast, and their evaluation is based on depth. One should use knowledge-based engines, that do something like we do ourselves.

I repeated Kai's experiment with opening analysis and got plenty of useful informations from a combination of older engines I know well (The Baron, Gandalf, Zarkov) when checking the lines by hand, and giving more time to engines. BTW, you can use Kai's test set to get an idea about which engine to use in which type of positions.

Knowing the good and bad sides of your engines is critical: The Baron is better in some strategic kind of positions, Zarkov is reticent to move it's pawns in the middlegame, Gandalf is weaker in simple positions but sees tactics faster, etc. Comparing the output is useful.

4. Playing against engines can be done in different ways: tactical games, games against weaker engines (or programs: I often use older programs from the middle nineties), playing resulting positions from the repertoire, playing with the possibility to see the engine output/without that possibility, etc. The key is to play at LTC and to mix the exercises. The result doesn't matter although it's better to win, of course.

I know that Shredder has a "rated games" function, so if you own Shredder, it's also a good way to train. I know that a few years ago you could try the GUI for a month. If it's still possible, it's a good way to start.

5. Improving your technique with an engine: I used Wasp 1.02 here, but you can try iCE or The Baron and analyze complicated endings. Here too, comparing different engine's output with your own ideas is critical.

Memorization: I used the free version of CPT once, but most of the learning is done in the analysis. If you are not a master, or if you don't chose highly tactical lines (and I don't recommend using such kind of openings for casual play and a level below or around 2000 Elo), programs directed on learning repertoire moves are mostly useless. I played the Simagin variation and got only twice to the end of the book line: and, surprise, I discovered that the line that was evaluated with += in the book, was, in fact, lost for black. Had I analyzed and played this position I would have known that.

Engines don't understand certain type of positions, of course, but they can be very useful. A combination of play and analysis, different kind of positions from different phases of the game, use of different engines gives solid results. The key is to work yourself the most you can and not to rely on engines to work for you, but with you.

[Ferdy]

I use engine to analyze positions and interpret its idea based from the pv. Then try to remember the idea when something similar position is encounted in actual game play or during analysis sessions. I usually do this when the position is quiet. As you increase analysis time, the engine may change its plan. Capturing this new plan is also interesting.

I don't add too much variations in my blitz repertoires, instead I look for patterns and take a note on that pattern.

The most difficult thing when you are on the board playing is to come up with a plan from a quiet positions. Then you may worry if the plan is good or not. You may then consult the engine later after your OTB game. Building these ideas/plans/pattern into your repertoire would be a good step to a stronger player I beleive.



I use CPT, chessok aquarium idea (correspondence), old and new bookup. But I so far I like the old bookup for its simple/clear interface.

[Ovyron]

I use engine to analyze positions and interpret its idea based from the pv.

In my experience the engine's PV isn't very useful, and one would get a more powerful idea on the same time if one made the engine play against itself from that position at lower depth. Because every move of the PV is weaker than its predecessor, while with a forced game all moves of the line turn out to be of similar quality.

[leavenfish]

Ferdinand Mosca - This is kind of what I do with CPT. I have basically 'two databases'

ONE: Deep lines. Some lines really do go deep before getting out of theory of course. This one I try to keep updated ever so often and is the 'master file'. A giant TREE you might say.From this file I have a second:

TWO: Shorter lines for constant training/memorizing so I am less likely to forget something I won't see very often or might see in blitz or quicker time control games. A SHRUB is how I think of it - shorter, wider.

With this second one I may only go...say, 13 ply with lots of branching points . At the end of each node I make comments about exactly what I should be trying to do from that position. That being easier to remember than actual moves...more practical for OTB play. This is great to review for OTB tournaments.

Along the way with each I like to run engines in multi-pv so that it can point me to situations where my opponent has more chances to go wrong or drift - blunders even. Having an engine spot these is helpful as you might recall in the heat of OTB battle 'I believe I have already seen why this odd move does not work'. It is a game of mistakes after all!

[Ferdy]

I use engine to analyze positions and interpret its idea based from the pv.

In my experience the engine's PV isn't very useful, and one would get a more powerful idea on the same time if one made the engine play against itself from that position at lower depth. Because every move of the PV is weaker than its predecessor, while with a forced game all moves of the line turn out to be of similar quality.

I understand what you mean by pv moves quality. But letting the engine analyze a position for a longer time and then interpret the pv is also good. While the engine is analyzing you write your interpretation based on pv. I generally like to record the idea than record the variations. This is mainly for blitz training.

[Ferdy]

Ferdinand Mosca - This is kind of what I do with CPT. I have basically 'two databases'

ONE: Deep lines. Some lines really do go deep before getting out of theory of course. This one I try to keep updated ever so often and is the 'master file'. A giant TREE you might say.From this file I have a second:

TWO: Shorter lines for constant training/memorizing so I am less likely to forget something I won't see very often or might see in blitz or quicker time control games. A SHRUB is how I think of it - shorter, wider.

With this second one I may only go...say, 13 ply with lots of branching points . At the end of each node I make comments about exactly what I should be trying to do from that position. That being easier to remember than actual moves...more practical for OTB play. This is great to review for OTB tournaments.

Along the way with each I like to run engines in multi-pv so that it can point me to situations where my opponent has more chances to go wrong or drift - blunders even. Having an engine spot these is helpful as you might recall in the heat of OTB battle 'I believe I have already seen why this odd move does not work'. It is a game of mistakes after all!

Deep lines are really good for correspondence plus you can do mini-max in CPT. Its a cool program.
Taking notes is time consuming but its rewards are lasting.

Multipv is one of the strongest feature of CPT that bookup is missing. I tried to convince the bookup author to implement it, but seemed like it was not in his priority list.

I developed a tool before where you try to play a move, not the best but not bad either but will make the engine (after that move) changes its root moves too often, meaning we may call it not an easy move. The idea is if engine often changes its root move during iteration, it is also possible that your human opponent may experience difficulties during his analysis. There are positions whose top 3 moves may just lead to a draw but each of those moves may lead to a different ideas, and there are ideas that are not easy to see. But of course this is only applicable when playing against human.

[leavenfish]

Deep lines are really good for correspondence plus you can do mini-max in CPT. Its a cool program.
Taking notes is time consuming but its rewards are lasting.

Multipv is one of the strongest feature of CPT that bookup is missing. I tried to convince the bookup author to implement it, but seemed like it was not in his priority list.

I developed a tool before where you try to play a move, not the best but not bad either but will make the engine (after that move) changes its root moves too often, meaning we may call it not an easy move. The idea is if engine often changes its root move during iteration, it is also possible that your human opponent may experience difficulties during his analysis. There are positions whose top 3 moves may just lead to a draw but each of those moves may lead to a different ideas, and there are ideas that are not easy to see. But of course this is only applicable when playing against human.

That would absolutely be something useful for preparing lines for OTB play! I wish something like what you developed were easily available...something engine developers could try rather than chasing increasingly meaningless elo. but endless tweaks are easier I guess.

I try to do that sort of thing manually - look for 'perfectly good' areas to move the game forward, where the opponent is presented with more reasonable looking choices where they can 'go wrong' and give me either easier play (in problematic openings) or better chances to get an edge. Being time consuming (but fun), I can only do that in fewer places.

I suppose one could do this in 'Correspondence' play as well...but that would be terribly difficult. My Scandanavian (Bg4 lines) or 1.Nc3 (lots of transpositional ideas!) approaches...don't translate well. But OTB, they can be VERY hard for my opponents to handle. Most my (engine allowed) Correspondence play is simply geared toward giving myself the chance to explore those risky lines more - winning or losing not being so important - looking for 'wrinkles' in established lines for OTB play being what I am after.

It is what I call The Lasker Approach to choosing moves for OTB play. He sometimes would chose moves that simply gave his opponent more chances 'to go wrong', when another move might have given him a slim edge but with fewer chances for Lasker himself to convert. Only...I want to do the prep before hand, since I am pretty good, but no Lasker!