My engine Spandrel shows regular alternation of scores during iterative deepening from the start position. Scores for odd depths are greater than for even depths, initially by about a quarter of a pawn, reducing to about one eighth.
I remember suggesting in that thread that a side-to-move bonus should be set to about half the average value of being able to play one move (about half the value of a tempo). Reason being that on average, side-to-move will get to play 0.5 moves more than the opponent before the game is over, so (all else being equal) if the outcome of the game isn't decided yet, side-to-move should have a slightly better chance of forcing a win.
micron wrote:My engine Spandrel shows regular alternation of scores during iterative deepening from the start position. Scores for odd depths are greater than for even depths, initially by about a quarter of a pawn, reducing to about one eighth.
Is the alternation a reasonable expression of white's first-move advantage?
Robert P.
Yes it is always an advantage to play first in the early part of the game. The way the score are progressing, it is also a sign of a bugless program. Probably also evaluation of pawn structures should be tuned and improved. Piece interaction to pawn structures should be added and tuned.
By adding a "side to move bonus" to your evaluation function, where the bonus could be about +0.10 ... +0.15 in the opening (maybe less?) and decreasing towards the endgame, your odd-even effect would certainly disappear.
Btw, regarding the PV example you have shown I'd like to add that the final PV of your program looks a bit unusual. The position after
is clearly much better for black, due to the weak pawn on d4 and other positional factors (in fact Black wins the pawn, e.g. Nf6 Be3 Bc5 -+). So there could also be another problem with your evaluation.