2 * Manual page - full usage documentation
3 * GTP interface documentation
6 * Further optimize board implementation, profiling fun
7 * Clean up GTP interface, allow custom GTP commands for modules
8 * gogui-friendly GTP interface
9 * Implement parameter setup over GTP (less important)
10 * Fix the build system to allow fully parallel build
11 However, revamp to something like cmake (or, ugh, autotools)
12 is not guaranteed to be appreciated.
15 * Improving Pachi's game analysis features
16 We provide just a few user-unfriendly proof-of-concept scripts
17 but it should be fairly easy to upgrade them to something
18 that creates a nice webpage with move-by-move statistics,
19 winrate evolution, pattern moves andwhatnot. CrazyStone stats
20 output may be used for inspiration, but we can take it further!
21 This could be done even if you are afraid of Pachi's codebase,
22 just using Pachi's output.
23 * Try to disable the bsize pattern feature
24 It just fudges the pattern evaluation since for most tactical
25 patterns fourth line vs. fifth line just doesn't matter. Maybe
26 its max should be 2 and maybe it should just be gone, needs
27 regenerating the pattern database and benchmarking.
28 * Develop dedicated playout handling for few common tactical situations
29 Monkey jump (and its followup sequences!), bent four in the
31 * Fix seki recognition to be stable
32 Try to research and fix most cases harmed by selfatarirate=100
33 or find another way to fix seki.
34 * Try to avoid using a hash table for 3x3 patterns
35 Instead autogenerate procedural matching code; may be more
36 efficient (the near-guaranteed L1 cache miss is fairly expensive).
37 * Optimizing our tree implementation for cache-efficiency
38 Statistics of all children of a parent node shall be contained
39 in an array of the parent node so that move evaluation during
40 the descent can access them sequentially in memory, instead
41 of walking a linked list. Pasky already tried once but it's
42 somewhat arduous and dull work.
43 * Clean up the is_private() hack in the distributed engine
44 We should simply check against a proper IP range ACL specified
45 as a parameter instead.
48 * Online Pachi game analysis/dump features
49 Make Pachi generate a webpage with *lots* of details after each
50 move while playing a game. This can provide game analysis info
51 for observer or casual opponent, but could also help with
52 debugging when just dumping stuff on stderr is intractable.
53 * Automated building of opening book
54 * Expanding and tagging the regression suite
55 Even better, create a nice UI for our users to contribute and
57 What about drawing testcases from GNUGo's regression suite?
58 * Implement Pachi support to fishtest
59 http://tests.stockfishchess.org/tests would allow crowdsourcing
60 Pachi parameter tuning.
61 * Split playout aspects to custom-stackable pieces?
62 * Port to Intel Phi (if we get the hardware :)
64 Some heuristics to test:
65 * Try to adapt and reuse GNUGo's pattern matching code/database
66 * Local trees (work in progress, no luck so far)
67 * Liberty maps (work in progress)
68 * Implement a tsumego solver and apply it once per playout (stv insp.,
69 see Eric van der Werf's PhD thesis?)
70 * MM local-based patterns in playouts (work in progress, no luck so far)
71 * Balanced local-based patterns?
72 * Killer moves (redundant to RAVE?)
73 * Reverse status learning
74 Run on game corpus. Start at final position, watch development
75 of status of all stones. The moment the final status and expected
76 status changes, analyze, especially if move choice differs. Use
77 learnt status-fixing moves in simulations somehow.
78 Tried to do this on Pachi-played KGS games; no measurable effect
79 (maybe too small sample).