Tournaments

  • The idea is to compare Milady versions to some dedicated computers. Why? Because I suppose they are better compared to humans. The strength of dedicated  computers is well established. For example the Wiki Elo List gives a rather complete overview of dedicated computers ratings, with different scales (Aktiv = rapid chess, Turnier = Tournament, SSDF rating, BT-2630, BT-2450, Colditz).
  • I have performed some tests to verify the scale given by Wiki, just by curiosity. Performing 20 games between Mephisto MM II and Mephisto Expert Travel Chess, (1min/move for Expert, Level 5 for Mehpisto MM II) gave me the following result:
  •  


  •  Some explanations on the above table : Score is the total score of the computer (including draws), rating is the initial rating of the computer (from Wiki Elo), Perf (Elo) corresponds to a very simple formula (62,5% of Perf = 100 elo rating difference, 75% of Perf = 200 elo rating difference,...) which is not accurate is the two opponents are very different in stength, Error Barre in % is calculated by the formula 40/root(number_of_games), Elostat means the Elo formula has been used for the calculation of the Elo (but not the Elostat programme, nor the Bayeselo of Remi Coulom. One difference is I consider that a draw is part of the performance as it is done in the official calculation rules). The results are very consistent with the Wiki Elo so that I can be satisfied with the Elo they give for this two machines.
  • Here you can find the pgn file of all the games I have recorded
  • I will add some other verifications tests in the future

2 comments:

Old Len Are Beautiful said...

Milady link is broken

Mauro said...

Hello.
Mee too... link are broken.
Can you repost please? I'm very interesting in this program.

Thx