[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Bug-gnubg] miscellaneous bugs and suggestions
From: |
Robert-Jan Veldhuizen |
Subject: |
[Bug-gnubg] miscellaneous bugs and suggestions |
Date: |
Thu, 05 Sep 2002 04:18:20 +0200 |
Hi team,
a whole list this time :-)
bugs:
- "alert! missed double (x.xxx)" gets inserted when copying data from hint
window
- rollout windows on my PC are often a bit too small (last two digits of
"cubeful" not shown) so I have to resize them.
- bug or feature: in money games using Jacoby with cube in center, gammons
don't count for cubeLESS equity, both with cubeless rollouts/evaluations,
and in the upper row when rolling out a cube decision (or doing a cubeful
rollout with cube centered). I think cubeless evaluations and rollouts
should ignore the Jacoby rule (after all it's cubeless) and just count
gammons (and play according to that!). For the rollout cube decision, even
if the cube is still centered now, it'll probably be turned in the future
so this seems more like a bug really. In fact it seems that winning gammons
on a center-cube with Jacoby is extremely unlikely with "normal" play, so
maybe just all gammons should count in a cubeful rollout, even when the
cube is still centered (although play should be according to the Jacoby
rule here!).
questions:
- if I use 3-ply lookahead and no 1-ply pruning, will the search space I
set be used for 2-ply, and half of that for 3-ply? That is what I'd expect,
but I've been wondering if that's what's really happening. Could it be that
the 3-ply search space is a quarter of what I set, as would happen WITH
1-ply pruning?
- for cube decisions, is the search space relevant? If so, how is it being
used?
- just wondering, is it clear to any (native) English speakers that a
"novice" is a better player than a "beginner", as by GNUBG's ratings?
suggestions:
- it would be nice if there was a quick way to see the MET in use. Perhaps
the "options" window could show that near the "load" button?
- I think the MET in use should be included in rollout results; at least as
an option. It can have a significant impact on results sometimes. And
perhaps the random generator and seed could be left out of the report as an
option then.
- perhaps some "direct access" 0,1,2 and 3-ply evaluation buttons in the
hint and annotation window, so one doesn't have to change the search space
for this (and change it back to its original setting even).
- with today's CPU-speeds, maybe reduced (33%) 3-ply evaluation would be
useful (for analyzing mainly), or even 33% 4-ply?
- maybe the "noise" settings could disappear from the GUI? Some users seem
to be confused about this and the few(?) people using it might have little
problems using the text command?
- after analyzing a match, when I want to do a better evaluation or a
rollout on a cube decision, the result is only shown (and counted in
statistics) for the player on roll (either the doubler or the taker). Since
all values are computed anyway, why not change the figures for both
players? Now I have to do an identical evaluation/rollout, taking twice as
much time. See next suggestion though:
- how about extending "rollout cube decision" so that one can also just
check the double decision (only upper rollout) OR take decision (only lower
rollout), instead of both? Sometimes it's just one side of the decision
that's really interesting, and this would be faster. Combining this with
the above suggestion, the user could choose to either rollout both
decisions at once and figures being updated for both players then, or just
for one player if he rollouts out for only one part of the decision.
- the tolerance for "world class ++"/"enormous" is set at 1.0, that seems
unnecessarily large to me: at 2-ply it is very unlikely that a play is
best, that is more than 1.0 worse than the best play at 0-ply, or 0.5 worse
at 1-ply. As a result, many plays that are downright awful will be
evaluated at 2-ply. Now the standard "world class" setting uses tolerance
0.16, which is ok but can be too small for some purposes. Maybe something
inbetween would be most useful, like tolerance 0.4-0.5. That'll leave out a
lot of bad plays and thus save time, while still finding the best move
probably >99.9% of the time.
- perhaps besides the full speed ahead "gigantic" setting, which is more or
less a non-restricted search space, maybe something inbetween would be
useful too. Like 32 candidate moves, tolerance ...well, 0.6-0.7 perhaps.
Some positions with many legal plays can have lots of close plays too, and
16 is sometimes just not enough.
Thanks all for your continous efforts on GNUBG!
--
Robert-Jan Veldhuizen