Ed Gilbert wrote:
I finished an engine match with flits this morning with pondering off, and kingsrow had fewer wins than usual. It could just be randomness and coincidence, but I'm wondering if pondering was the cause of flits making bad moves in the endings? Maybe you will notice this also.
I finished an engine match with Flits with pondering off today. With Flits not pondering and Maximus now using 8 (hyper)threads instead of 6, Maximus now did (6%) better, which makes sense. More glitches than usual, however!
Maximus vs. Flits +20 -23 =113 *2
after correcting 11 glitches and 2 unknowns:
Maximus vs. Flits +9 -23 =126 45.6%
Time controls were: Maximus 5 sec/move and Flits 75 moves in 5 min (equal). Games are stopped by Maximus when it sees a database win or loss, or else when there are 6 pieces or less on the board (database draw), or else after 75 moves (unknown).
All glitches follow the same pattern: they appear in positions with 6 pieces (or less) with Flits to move, after Maximus made a forced (re)capture (i.e. an immediate DXP response). The capture made the position a database draw position, but Maximus first sends this move and stops the game after Flits's next move - and then it often goes wrong. For instance:
Maximus - Flits, game 17
black to move, after 52.36-31 26x37 53.32x41 ...
Flits now played 53. ... 21-43? The other 10 glitch cases are, except for the position, identical.
What I can do is try to stop the games earlier when approaching a database draw. (Maybe an extra delay before sending an immediate MOVE?) See if they go away then. Pondering off seems to make the problem worse, but it can also be randomness.
I'm interested in how many glitches and when they occur other programmers who played matches against Flits can report.
BTW Ed, there is a tiny omission in the Flits DXP server: when the DXP partner acts as match initiator and sends the GAMEREQs, the Flits opponent names in ...\Flits DXP server\dxpgames.pdn are empty. As Maximus also stores the games this is not a problem for me.