How strong is Maverick 0.2?

Maverick 0.2 has been out a couple of days now.  I’ve run some test games and I really quite surprised at how strong it seems to be.

Fruit 1.0:

The first test was against Fruit 1.0.  I played this match using Shredder’s GUI. As I expected Fruit won.  I really was dazzles by Fruit’s silky smooth search as it carved it ways through each ply (especially in the endgame).

  • CCRL Rating = ???? (between 2300 and 2400)
  • Time = 1 minute plus 1 second per move
  • Result for Maverick: Wins = 24, Loses = 63, Draws = 13
  • Percentage = 30.5%
  • ELO Difference = -146

I was pleased with this result.  However, it would seem Fruit’s strength is not well established.  As far as I can tell it us somewhere between 2300 and 2400 ELO on the CCRL scale.  This would put Maverick 0.2 between 2150 and 2250 ELO.

Phalanx XXIII:

The second opponent was Phalanx XXIII.  Another classic engine.  As I watched them play it is clear Phalanx has much more knowledge than Maverick.  Yet Maverick did well to win some game based on its speed.

  • CCRL Rating = 2387
  • Results for Maverick: Wins = 36, Loses = 49, Draws = 15,
  • Percentage = 43.5%
  • ELO Difference = -45

Once again I was pleased.  Based on this result Maverick is over 2300 ELO, which is probably too high but let’s see.

And remember, at this stage Maverick is lacking some standard features:

  • No mobility in the evaluation
  • No passed pawn code
  • No king safety
  • No pawn structure code
  • No late move reduction
  • No Internal Iterative Deepening
  • Lots of endgame evaluation missing

So I’m encouraged by Maverick’s strength.  If you run any sort of ratings test I’d be interested in the results.

I will not release another version until it has gained 100 ELO.  This may be some time.

  • Hi Steve, congratulations to the new version.

    I did a quick (and unfair) test against iCE. 10 games at 1/1 where iCE did win all games. iCE had an advantage as it could use its opening book. Your engine is very fast, about 3 times as fast as iCE when looked at the raw nps data and it runs stable. I had no issues. As your evaluation is still a bit simplified and iCE searched a few plies deeper Maverick was just maneuvered out over time. Anyway your version looks very promising, keep it up !

    Thomas…

  • Steve Maughan

    Thanks Thomas! iCE is awesome and much stronger than Maverick 0.2 – at least 300 ELO. But I am pleased with Maverick’s speed considering it is doing SEE and checks in the qsearch. I’m now focusing on the evaluation. Once I have put in place the basics (mobility and passed pawns), I’ll probably take a similar approach to you and start with the endgame knowledge.

    Steve

    P.S. I expect the nps to fall by at least 30% once the eval is beefed up.