Mida v1.2 released

[Jack03]

I'm happy to announce the release of Mida v1.2, which should be aruond 2350-2400. I thank Graham Banks and Gabor Szots for testing the previous version and making me realize it wasn't where I thought it was. I worked hard and introduced some novelties that should bring this version where I hoped.
I did some proper testing this time, and these are the results. Against v1.0, I played a 60 game match in 30+0,4 seconds format, with these outcome:

Score of v1.2 vs v1.0: 39 - 11 - 10 [0.733]
... v1.2 playing White: 20 - 6 - 4 [0.733] 30
... v1.2 playing Black: 19 - 5 - 6 [0.733] 30
... White vs Black: 25 - 25 - 10 [0.500] 60
Elo difference: 175.7 +/- 91.4, LOS: 100.0 %, DrawRatio: 16.7 %
60 of 60 games finished.


Just to make sure, I also played some games against Hopper, an engine on that 2350-2400 ELO range in CCRL blitz, and the results were pretty drawish, confirming my hyptothesis, therefore I stopped the testing early . I stoppped the testing at 43 games with these results

Score of v1.2 vs Hopper: 15 - 15 - 11 [0.500]
... v1.2 playing White: 10 - 8 - 3 [0.548] 21
... v1.2 playing Black: 5 - 7 - 8 [0.450] 20
... White vs Black: 17 - 13 - 11 [0.549] 41
Elo difference: 0.0 +/- 93.0, LOS: 50.0 %, DrawRatio: 26.8 %
43 of 100 games finished.

[Gabor Szots]

Good to hear you made progress. I have just submitted my first bunch of games for v1.1 so don't be surprised if 1.2 will have to wait. I may skip it altogether if you release a newer version in the meantime.

BTW, 60 self-play games won't prove very much. 60,000 games, that's something.

[Jack03]

Thanks as always. About the number of games, I don’t think I have the computing power necessary for 60000 games, unless I use some ridiculously small time controls. If that’s the case, can I ask what time controls do you use for such a big number of games?

[Gabor Szots]

Thanks as always. About the number of games, I don’t think I have the computing power necessary for 60000 games, unless I use some ridiculously small time controls. If that’s the case, can I ask what time controls do you use for such a big number of games?

I was joking. I play only the games that are submitted to the CCRL. And you know the time control, it is 2+1 (or equivalent).

But to say you have properly tested Mida by playing 103 games is a huge overstatement. Look at our lists: to achieve an estimation error margin of +/- 20 we have to play about 1000 games per engine.

And be careful with self-play testing: experience shows that the difference you get by self-play is about 50 % bigger than the difference you would get against a large pool of various engines.

[Jack03]

All right, thanks