A Brief History of Draw Avoidance

Have you ever seen a position with a book value of +0.00 that turned out to actually be a win? I have. This happens when a move off the draw tree that is considered by the program to be of negative value, is actually a win when solved. This happens most often when the move has a negative value is very close to 0.
There are two ways to deal with this problem. Suppose the winning move is scored as -0.37. You can set the randomness of the selfplay to 1 disc in the hope that this -0.37 move will get selfplayed. This may not be so bad if the randomness is small, and the winning move is initially scored as close to 0 by the program. What happens if you suspect the winning move may have a score beyond, say, -3.00? Your chance of hitting the right line gets smaller and smaller as your randomness gets set higher and higher, so opening randomness becomes less and less helpful the further out the winning move's score lies. Additionally, the use of opening randomness will sometimes result in previously played games being replayed, which is a waste of time.
So, the other way to deal with this is draw avoidance. If in your opening book you assign a score of negative infinity to all actual draws, all solved draws will be avoided, and the deviations scored closest to 0 will be selfplayed in the next game (as it has now become the highest scored avaiable option- remember, all drawing moves are now negative infinity).Thus lines that are the most likely to be wins will be directly selfplayed. This is a systematic way of correcting false draws in an opening book. You also do not have to make a subjective judgementabout where the winning move may lie, if there is one. With enough time, draw avoidance can solve the opening for you, for one color anyway. If you are using white draw avoidance, black will either hold the draw, or the best losing sequence for black will be found. (In effect, selfplaying with draw avoidance is a form of selective solve)

"Black score" and "White score"

These are the names that I took from ntest's code. The "black score" is the value of the position, if draws are treated as wins for the next color to move. (Notice on the listbox, the "black score" of each move is the "black score" *AFTER* that move is made). "White score" is the score if draws are treated as losses for the next color to move. (Likewise, in the listbox, the "white score" of each move is the "white score" *AFTER* that move is made). Confusing? Yeah I know.