--- In abalone_prog@yahoogroups.com, david.malek@... wrote:
>
> Hi Sébastian,
>
> Welcome to the Abalone world :)

Thx :)


> >Engines
> Aba-pro is a powerful engine, I think Tino has done a very
impressive work. I
> took me ages to beat this engine with MLA. I'm sure abapro is faster
in node
> evaluation than MLA.

I'll look at both. I still havn't asked Tino for AbaPro, but I'll do
so once I find the time.


> > I am able to reach higher search depths than current Abalone progs (10
> > ply I was told in case of Aba-Pro).
> Exploring the game, and paticulary the tree cutting issues, legal moves
> generation, position evaluation will lead you to the max depth
you'll be able
> to reach lol (for example, ProbCut doesnt seem to work and Aba-pro's
heuristic
> has been proven wrong in its expansion strategy by MLA is some
cases, other
> example : bitboards seem to lead nowhere).

I took a look at bitboards, but I think that - while certainly doable
- they won't give too much of benefit.
IMO, search depths are only substantially influenced by two factors
(taking basics like a fast search algorithm, hash tables, etc. for
granted):
1) the quality of move ordering
2) how to cut/reduce/prune moves/sub-tress that are a waste to be
looked at (or rather: that the benefits of not searching these is
higher than the drawback of an unlikely move missed)


Number 1 is of course a free-lunch exercise, but at some point a
ceiling will be hit with little further possible improvement.
I think that the second route is still not explored as much as it
could be (not even in the chess world). It's of course an exercise in
testing whether a particular reduction/extension or prunging is
benefitial or not...

To put it in my favorite way of thinking, there is no search
depth-ceiling that cannot be reached. But there is a ceiling in nodes
that can be analyzed. The problem always is where to spend those nodes...


> > field, pursued without the required professional approach.
> Lol, do not underestimate Peer, Tino and others :) Particularly
their knowledge
> of game programming (for both search and evaluation lol).

:) Now that i reread that statment it sounds quite arrogant. Wasn't
meant the way you understood it though.
The lack of professional approach didn't refer to the qualitiy of
their/your programming.

It referred to the substantial lack of :
1) specs and agreed upon formats (move-formats,position-formats,etc.)
2) uncontended and unamibgious notation!!
3) unifying GUI's and ability to easily compare and contest engines
4) test-suites (for positional/tactical analysis)
5) rating-lists
6) discontinuity of strong programs

and other such things.
It's quite probably due to the (small) size of the field. But (IMO),
if that could be changed, the whole field would become broader and
better.
It's an issue of how much time is spent once to achieve these results
and how much is saved later on, if once done.

> > I also want to provide a GUI that'll include a lot more powerful tools
> > for analysis and game/database management.
> There is a game file format in MLA, and some tools (like a
Macromedia Flash game
> player for your web pages), feel free to contact me for further
info. It's the
> same concerning API/engine communication, Peer is also using the AbNet
> protocol.

I'll post here soon, regarding GUI/engine communication.


> > What i noted most about Abalone (in contrast to chess) is the lack of
> > information even regarding such simple things as rules/notation.
> Rules and notation are around, MLA allows both notations. To me,
theoretical
> knowledge of the game is an important issue. From the opening book
automatic
> generation work i've done so far (no yet embedded in my engine) this
is a
> tricky issue. But knowledge generation still looks possible to me.


Regarding notation, that ain't an issue for the GUI. You just give the
user the option to use whatever sound or crappy notation the user
wants to use.
It is more of an issue in common protocols, file-formats and the like.
Especially broadside move notation, where I've seen so many variants
that it is hard too keep track of them.


> Note that my MLA work is in standby for the time being, these days,
i'm more
> focused on my girlfriend, my job and vacations (i'm back from mexico
today and
> will run to kenya in july).

:)

> Good luck, build competition to Nacre, Abapro and MLA :) Feel free
to share your
> ideas, ask for any info / litterature that could help you.

I will.
Search algorithms, I'll most likely be able to share a lot of
knowledge, while I might have questions about what has already been
tried in evaluation.

> Promise : If you beat MLA i'll put my hands in the engine again lol.
2.5 is the
> current version, if you beat it. 1.8 is the one to beat, it's the
won that won
> against abapro.
>
> David.

:)
Better remember that promise, ;)


Regards and thx,
Sebastian Leibnitz.