Rein Halbersma wrote:
Ah good point on full WL vs full DTW dbs. My opinion is that full DTW (or smaller DTW + quick search with correct win distance) is much more valuable than full WL, especially since smaller WL + quick search are good alternative.
DTW is obviously very valuable for international Draughts game where loops are possible. As an example, in order to win, it does not help very much to have a complete picture of the 5Kx2K winning positions We need really a complete picture of the DTW for each 5Kx2K positions in order to reach really the win.
BT is quite different because it could not exist loops. In such game a full WL is suffisant; of course DTW allows to highlight a smallest path to the win but this is not essential.
BTW DTW is a pure computer criterion which often makes no sense for a human.
Let’s take the two following positions:
In the first one I guess that the DTW is quite high but for a human point of view there are no doubts at all : white wins full stop.
In the second one I think the DTW is far less but now the human player needs a very long and difficult analysis to decide between a win or a loss.
As you see the DTW may not be the best criterion for a human. I am not able to propose a objectively better criterion but, as human, I prefer a non 100% sure approach like the following: suppose you have a good eval function (like the scan one ?). You can define a position as “almost win” is the eval function tells you that the advantage is superior or equal to one man. With this definition you can consider the DTAW (distance to almost win) concept.
With DTAW the first position above is already a “almost win position” and DTAW = 0 and this sounds quite satisfactory for a human.
On the other hand DTAW is obviously far higher in the second position and this makes really sense for a human, doesn't it?