Home MadLab Info Performance
MadLab performance

In general, the more tactical a problem is, the better the chances are that MadLab can to solve it. MadLab is not designed for solving slow-paced strategical problems where the outcome depends upon connectivity, eye-spaces etc., and where the issue is the long-term fate and stability of groups of stones.

MadLab is not good at solving life & death (tsume-go) problems, for which a specialized solver like GoTools by Thomas Wolf is recommended (GoTools is also a part of SmartGo). The following figure illustrates the difference between a life & death solver like GoTools and a tesuji solver like MadLab. GoTools solves all kinds of enclosed problems; even if they have very high lambda-orders (i.e., are of slow-paced, strategical nature). For instance, GoTools may spot if the defender side has no room for making two eyes, even if it would take the attacker a large number of moves to actually remove the stones from the board (indicating high lambda-order). In contrast, MadLab performs best for lambda-orders n=4 or lower, but the problems do not need to be enclosed. 

Figure: Difference between a life & death solver (GoTools) and a tesuji solver (MadLab)

The uncovered area in the above figure (open space, high lambda-order) corresponds to, for instance, life & death problems with holes in the enclosing group of stones, or semeai problems (these are not 'enclosed' in the GoTools-sense that one color encloses the other). Semeai problems typically have high lambda-orders. No commercial product is available as far as I know, but Martin Müller has done a lot of work in that field.

 

Performance on Davies Tesuji

   

MadLab has been tested on 129 problems from the well-known book Tesuji by James Davies. The problems are from chapters 1-3 and 5 (i.e., "Reading", "Capture the cutting stones", "Amputate the cutting stones" and "When liberties count"). Chapter 4 ("Ko") is excluded from the test, even though MadLab can solve ko problems. Of the 129 problems, 10 are not translatable as capture problems or have tsume-go or semeai flavour (defined here as lambda-order n=5 or higher). The rest are solved by MadLab, however a few of them take a long time to solve. This is typically because the defending side has a non-working counter-attack that confuses MadLab. In a future MadLab version, such counter-attacks will be decomposed into separate sub-problems, and at that time such problems should hopefully solve much faster.

MadLab has not yet been tested on the remainder of the Tesuji book, focusing instead on the fast-paced tactical problems in the first part. The program also performs well on tesuji problems from Kano's Graded Go Problems for Beginners vol 1-4.