Zevra time management ?

[tissatussa]

i was testing your new 2.3-rev348 in CuteChess, 3 min per player + 2 sec bonus .. Zevra plays Black : 1.e4 Nc6 2.d4 h5 ?! after 6 sec (!?), but when it has more time per move, it finds the normal 2...e5 best move after 17 sec ..

how is determined to stop searching ? A next depth could give a better move !
how does Zevra handle my 3 min / 2 sec ?

[sovaz1997]

in Zevra simple time management - Zevra allocates time per move depending on which stage of the game + adds the increment. If time runs out, the search process is interrupted. Zevra can indeed make strange moves - this is the result of tuning the parameters of the evaluation function. I will increase the sample and this will allow us to refine the parameters. Perhaps Zevra will stop making moves with h-pawns at the beginning of the game)

[sovaz1997]

I'll think about adding a thing to the time manager that will allow it to count to the desired depth, if, according to the calculations, it will be worth it

[tissatussa]

i guess the search process can be interrupted, but then Zevra should give the "best move" of the last fully calculated depth .. this is my insight, but i'm not into engine programming ..

[sovaz1997]

Yes, if I notice this behavior, I will try to fix it. In any case, I re-trained the position assessment and now there will be no ridiculous moves in the opening :) Thanks! I leave the issue open for now.