multi-pv tournament?

[clumma]

Has a tournament ever been tried with multi-pv enabled (say, = 3)?

Many humans use engines for analysis with multi-pv on, but how do we know strongest engine in conventional matches is still strongest when running multi-pv?

[Engin]

yes would be interesting too, but the strongest chess engines are stockfish, komodo, houdini, so many of the people are using this engines to analyze his games.

[leavenfish]

I'm afraid I do not understand the question. It is not possible.

[clumma]

It's true that a supervisor program of some kind would have to make moves on a certain schedule (10s per move for example). Hardly impossible.

[petero2]

Has a tournament ever been tried with multi-pv enabled (say, = 3)?

Many humans use engines for analysis with multi-pv on, but how do we know strongest engine in conventional matches is still strongest when running multi-pv?

I know this is not really what you asked for, but it is related.

I modified texel to be able to use multi-pv mode when playing games. I played some matches using this texel version and different multi-pv and time control settings.

Code: Select all

Player 1           Player 2           Elo diff
MPV     Time       MPV     Time    
2        6+0.06    1        6+0.06    -52
2       12+0.12    1        6+0.06    +112
1       12+0.12    1        6+0.06    +144
2       12+0.12    1       12+0.12    -40
4       12+0.12    1        6+0.06    +19
5       12+0.12    1        6+0.06    -19

"6+0.06" means 6 seconds per game + 0.06 seconds increment per move, so these were hyper bullet games. The number of games in each match was at least 4000.

In summary, if you use a multi-pv value of 4 or 5, you need about twice as much thinking time to get the best move to be equally good as when not using multi-pv.

The above conclusion is of course not necessarily valid for other engines and for longer thinking time.

[clumma]

Very interesting, Peter. Thanks for sharing your results!

[yurikvelo]

I'm afraid I do not understand the question.

Question is: "If engine_A_ELO > engine_B_ELO" in singlePV mode, does it guarantee "engine_A_ELO > engine_B_ELO" in multiPV=3?

In other words: can multiPV hurt one engine more than other?
Is ELO difference between A and B the same in multi=1 and multi=3 mode?

Do we need to measure strength also in multi>1?