All the data for cgos is available as files zipped up by month. So
any kind of study is possible.
- Don
On Wed, 2007-01-31 at 10:44 -0500, [EMAIL PROTECTED] wrote:
>
>To gain some intuition re. what level of intransitivites are
> present on CGOS, it would be interesting to see the
On 1/31/07, Tapani Raiko <[EMAIL PROTECTED]> wrote:
> I imagine that the most significant intransitivity would be would be in
> relation to the bots (principally GnuGo?), because some players have
played
> dozens (maybe hundreds) of games against these bots and their playing
style
> is likely to
> Would the results of kgs (or similar) games being appropriate if one
> considered only un-handicapped games?
Yes, I think so. At least when considering only those who have played lots
of games against lots of opponents.
> I imagine that the most significant intransitivity would be would be in
On 1/31/07, Tapani Raiko <[EMAIL PROTECTED]> wrote:
Even if each player's performance is asymmetrical but identical, the
difference of performance becomes symmetrical again. But still,
intransitivity can be seen from results of matches. If one has enough
results of N people playing against each
To gain some intuition re. what level of intransitivites are present on
CGOS, it would be interesting to see the cross-scores of all (or some...) of
the bots in a big table. A less obvious refinement is to add color coding to
make it easier to read. Here's an example from the game of C
> If instead of assuming a player's performance follows a symmetrical
> probability distribution function (the normal in the case of Elo),
> assume the p.d.f is asymmetric, e.g. triangular. If player's have
> p.d.fs which are increasing, decreasing or rectangular, wouldn't
> that produce intransiti
Don Dailey wrote:
It would be interesting if it would be possible to construct a 2
dimensional model statistically. A 2 dimensional system would not
be a perfect fit either, but would simply be a better approximation.
And another constraint is, it must be supported by evidence. I think
this p
Nick Apperson wrote:
> There are certain times when this technique is highly useful. ...
> imagine a board with two walls down the middle bordering on each other
I agree. We have to divide the board to create strong programs!
But division is a very complicated subject. In the isolated areas UCT
Vlad Dumitrescu wrote:
Unfortunately, having more than one dimensions makes comparisons
impossible - if an ordering relation is defined over the domain, then
this domain is "one-dimensional" with regard to that relation.
In other words, one can't compare vectors, just scalars. So the
multi-dime
On Wed, 2007-01-31 at 04:00 -0800, Dave Dyer wrote:
> >
> >Of course, everything depends on how you can deal with the boarders - how
> >about some monte-carlo-simulations over the possible boarder-configurations?
>
> My thought is that one thing you could easily get from the rollouts
> is a good
My basic idea, which is undeveloped is rather like this - you partition
all players into 2 (or more) broad styles. One of the variables in the
rating
tells you how much (from 0 to 1) he plays at one extreme. The rating
system itself somehow determines which style you are and it's an
abstract
qua
I have experimented a little with getting moves that are likely
to be best first in the list - and in come cases giving them
extra weight so that quickly get expanded early. However, I
never showed a measurable improvement but I didn't stick with
it long enough to develop the idea fully.
My ba
>
>Of course, everything depends on how you can deal with the boarders - how
>about some monte-carlo-simulations over the possible boarder-configurations?
My thought is that one thing you could easily get from the rollouts
is a good estimate of the status of each string of stones currently
on th
I feel that what we need essentially is a set of functions that tell us
expected winning percentages with certain matchups. In an extreme example,
we could imagine 3 rock, paper, scissors players. One always plays rock,
one always scissor and one always plays paper. In this case, we would be
ab
Funny thing, I also just thought about this as a friend of mine had an
idea similiar to Dave's.
I guess it might be a good idea to make your zone-partitioning (or
zone-merging, when you start from 1x1-boards) dependant of the current
board configuration. That is, some clever algorithm (probably
Quoting Vlad Dumitrescu <[EMAIL PROTECTED]>:
I wonder if anybody has some data (or a theory) about how valuable a
good preordering of the children of a node would be, when UCT. This
might take the form of having nodes start with non-null values for the
'won_simulations' and 'played_simulations'
I'm working now in a similar idea.
As yours, it will play only in one zone using MC. It will start on 7x7
sub-boards but they will grow once they become full of stones.
I will normalize the sub-board results using its area. It will help me to
compare the different sub-boards.
Once all the sub-b
Hi all,
I wonder if anybody has some data (or a theory) about how valuable a
good preordering of the children of a node would be, when UCT. This
might take the form of having nodes start with non-null values for the
'won_simulations' and 'played_simulations' fields.
Or maybe this could work only
Hi,
On 1/30/07, Don Dailey <[EMAIL PROTECTED]> wrote:
It would be interesting if it would be possible to construct a 2
dimensional
model statistically. A 2 dimensional system would not be a perfect fit
either,
but would simply be a better approximation.So in some way a players
"strength" c
19 matches
Mail list logo
