Observation on positional tests

Answer

Question: - I "routinely" let the new version of Fritz carry out the BS-2830 test. Whilst the previous version achieves 24 out of 27, under the same conditions the current version only manages 20 of 27?! Although it is clear to me that the test is not a “panacea”, it does not even seem like a step in the right direction.

Answer: Using e.g. the simple positional test you cannot “measure” a parallel engine. You have to repeat the test many times in order draw any conclusions. A single run is of no value. 10x is better, 100x offers the user reliable data. Tactic tests also do not offer any real indication of playing strength. These tests only measure how fast the position can be solved. In order to say anything about tactical ability, a statistically significant number of positions are required. We recommend that a test with less than 1000 positions offer little value when evaluating an engine.

Tags
Created on
23.09.2016
Rating
Feedback

Back to List