HAL9000

HAL9000
"It just isn't conceivable that you can design a program strong enough to beat players like me."

March 11, 2015

Rapidroid gets wider, if not deeper!

New and updated engines come falling on our heads everyday. These days are as sad as happy. Weird statement, huh? Indeed, it's fun to see the pool gets bigger but you've got to face the truth that every update needs a rework of the past games and standings.

Every tester has its own methods to deal with updates, Some accumulate new versions over olders and keep old games just the way FIDE treats human players and some others start from scratch after each update, erasing previous versions and games, just the way a gardener deals with the grass generations spilling over and over. You just cut'em and wait for new ones.

A third method is to never remove any previous result and keep all versions together, like does CCRL. That ends with thousands of versions together, a much bigger database and requires special care for avoiding games between versions of the same engine, in order to prevent rating distortion

I'm simply going for the start-over method for Rapidroid, meaning that in time many games will have be thrown away and replaced with new ones. A little bloody, destructive but simplest way.

As of today, below engines have been certified good for tournaments after last Rapidroid release. They currently play gauntlets to find a place in the next ranking. I hope to submit it by the end of next week with about 2000 games total. The main harm with so frequently updated engines is that the number of games played per engine can't increase as quickly as expected. When there's no updates, we can proceed to play more games per engine to shrink the error margins. When updates are raining, it's unavoidable to spend time to refresh everything and delaying next rounds. Wider population means wider margins.

Finally, new entries will be:
* Joanna (weakest engine ever?)
* Mango Paola
* Murka (Unexpected puncher! A good one...)
* NGplay
* Protector (Looks much weaker than on a PC)
* Slowchess
* Trappy Beowulf
* Vajolet
* Colossus C64
* Colossus C64 200x (100x seemed not enough to me lol...)

Updates:
* Sayuri 2015.03.02
* Sting 4.8.4

That means the next release will issue 104 + 10 = 112 engines at least.

1 comment:

Gurcan Uckardes said...

News on March-13:
* Protector gauntlets are canceled for unstable terminations. Unfortunately, the engine is not stable.
* Pawny, Ifrit, Ecce, Clubfoot, Smash all look good for tourneys.
* Octochess r4995 will replace r4984 (Better fixed tpm mode, multicore, no wekaer than r6656)
That makes 104 + 14 = 118 engines expected in next release.