Weakest engine with multi-PV

[perejaslav]

Which is the weakest UCI chess engine with the Multi-PV support?
I look for it, help me, please!

[hgm]

I could make a multi-PV version of Fairy-Max for you, if you wanted. (Would be slightly above 2000 Elo, on the CCRL scale.) But Alas, it is WinBoard, not UCI.

It would be of some interest to do this, because then it would also be the strongest engine to support multi-PV in Knightmate, Gothic Chess, Berolina Chess, Courier Chess, Shatranj, Cylinder Chess...

[perejaslav]

It would be very nice from you to compile such engine but I'm interested only in UCI module.

[hgm]

Yes, that is what I was already afraid of. Then I cannot help you, as all my engines are WinBoard.

As for a multi-PV version of Fairy-Max, I guess I'll make that anyway. I already patched it yesterday night for printing a full PV (based on the triangular-array method), which seems to work fine.

[Kempelen]

I could make a multi-PV version of Fairy-Max for you, if you wanted. (Would be slightly above 2000 Elo, on the CCRL scale.) But Alas, it is WinBoard, not UCI.

It would be of some interest to do this, because then it would also be the strongest engine to support multi-PV in Knightmate, Gothic Chess, Berolina Chess, Courier Chess, Shatranj, Cylinder Chess...

Hi H.G.:
How exactly multipv is implemented? does it reduce playing strenght?

[hgm]

When you request more than one PV, it will reduce playing strength, because you spend extra time on evaluating second-best moves, which goes at the expense of the time you spend on evauating the best move. And this time is totally wasted, as you will not play second-best move anyway.

But having the ability to calculate multiple variations in itself does not put a significant burdon on the engine when you ask for only one PV.

The way I wanted to implement it is do a normal alpha-beta search, but in the root, I do not increase alpha to the score, but to score - MARGIN, where MARGIN can be set to the user. This will automatically recover all variations that have a score within MARGIN of the best. By setting MARGIN=0 you will recover the normal behvior.

You can also put a limit on the number of variations; in that case, if the user asks fr, say, 3 variations, when there happen to be more than 3 within MARGIN of the best, you can increase alpha to the score of the third, when that is larger than bestScore - MARGIN.

[jesper_nielsen]

Which is the weakest UCI chess engine with the Multi-PV support?
I look for it, help me, please!

Ok, I will start the bidding!

Pupsi has MultiPV support. Anyone lower?!
Pupsi is currently around 2375 Ruel on the UCI Engines Ligue list.

By the way, in Pupsi, you can reduce the playing strength using the Hyatt/Crafty method by setting the option "SkillPercentage" to a number between 0 and 100. MultiPV works with any SkillPercentage setting.

Kind regards,
Jesper

[perejaslav]

Which is the weakest UCI chess engine with the Multi-PV support?
I look for it, help me, please!

Ok, I will start the bidding!

Pupsi has MultiPV support. Anyone lower?!
Pupsi is currently around 2375 Ruel on the UCI Engines Ligue list.

By the way, in Pupsi, you can reduce the playing strength using the Hyatt/Crafty method by setting the option "SkillPercentage" to a number between 0 and 100. MultiPV works with any SkillPercentage setting.

Kind regards,
Jesper

I've found the weaker one Glass 1.3

Let's compare:

Code: Select all

  42 Pupsi2 0.08                 2376   37   37   258   46%  2405   28% 
  43 n2 0.1                      2360   19   19   964   51%  2352   31% 
  44 BBChess 1.3b                2350   17   17  1147   50%  2348   31% 
  45 Delphil 2.3                 2350   27   27   458   48%  2361   31% 
  46 Nejmet 3.07                 2343   10   10  3616   51%  2331   29% 
  47 Philou 3.0.0                2336   21   21   770   50%  2336   32% 
  48 Dragon 4.6                  2328   10   10  3852   50%  2325   29% 
  49 Counter 1.1                 2326   15   15  1401   50%  2321   38% 
  50 Snitch 1.6.2                2325   13   13  2083   49%  2329   27% 
  51 Abrok 5.0                   2324   10   10  4003   49%  2326   26% 
  52 Gaia 3.5                    2320   11   11  3278   51%  2310   23% 
  53 Sjeng 12.13                 2314   10   10  3950   51%  2306   25% 
  54 Diablo 0.5.1                2314   11   11  3094   50%  2308   28% 
  55 Patzer 3.80                 2305   12   12  2695   51%  2294   26% 
  56 Arion 1.7                   2300   11   11  2964   51%  2289   29% 
  57 Muse 0.899b                 2299   11   10  3364   50%  2294   27% 
  58 Queen 4.03                  2294   14   14  1802   51%  2287   29% 
  59 LittleThought 1.04          2293   33   33   336   44%  2344   24% 
  60 GreKo 6.5                   2289   20   20   852   48%  2306   28% 
  61 Glass 1.3                   2287   24   24   634   47%  2305   24% 

Info been taken from Statistiques complètes des Engines
My win so far?

[hgm]

I know this is not what the original poster is looking for, but just in case there is any interest:

A version of Fairy-Max that supports multi-PV can be downloaded from http://home.hccnet.nl/h.g.muller/multimax.zip . Through the engine option "MultiVariation Margin" you can set the score difference for which you want the variations printed (in centiPawn, default 0). So when you set it to 30 cP, it will print (at any depth) all variations that have a score within 30 cP from the PV. (And possibly some more, if it does not find the PV first time.)

So you can now have multi-variation in the variants normal, gothic, capablanca, berolina, knightmate, cylinder, courier, shatranj.

[tmokonen]

I think the answer to this question might be Firefly.

Firefly supports MultiPV, and is weaker than Glass. Firefly has a rating of 1983 in the UEL rating list.