27 January 2020

TCEC PCT Handicap Events

In last week's post on the engine-vs-engine tournaments, TCEC S17 L2, CCC12 Rd.3 Underway, I neglected to mention a series of small events that took place between the TCEC Qualification League (QL) and its League 2 (L2). These were five events with names like:-

S17 - Stockfish 1pct vs L2 Test 1 [through 'Test 5']

L2 has since finished and the TCEC is running a new series with similar names:-

S17 - Stockfish 3pct vs CPU L1 Test 1 [currently 'Test 2']

What are these 'pct' events? The site's help explains:-

!pct • tcecSF 0.3%, 1%, 3% and 10% are time odds games: tcecSF has only that amount of time compared to opponent. Both main time and increment is scaled down to given %. Hash size and Move Overhead is also scaled down.

!10pct • Only 10% of thinking time: 30min+5s -> 3min+0.5s; 45min+5s -> 4.5min+0.5s; 1min+1s -> 6s+0.1s

[...]

'Time odds'; OK. 'Hash size scaled down'; OK. 'Move overhead scaled down'; what does that mean? A forum post from a few years ago, The truth about time losses (groups.google.com/forum/fishcooking; September 2017), has a long discussion of the parameter and explains:-

Basic flaw is not properly taking into account future time losses due to communication delays (the thing that Move Overhead was supposed to compensate for).

In other words, 'Move Overhead' accounts for administrative tasks involved in the engine's communications with its environment; OK. How are we supposed to interpret the results of the seven 'pct' tournaments I mentioned earlier? TBD.

No comments: