"I also think match winning chance is anIf you throw a biased coin (lets say 0,52 head) you don't agree that the probability of winning betting on heads is a function of throwing the coin only once or best of 3 or best of 100?
"increasing function of match length for
"a certain PR difference.
"However, The PR itself is dependent onI have to commit that this sentence doesn't make much sense to me, maybe some context is missing. What might be meant is that the longer the match is, the higher the probability is that the measured PR is close to the real PR. One medium error in a 1-
"the match length.
MK schrieb am Dienstag, 6. Juni 2023 um 11:40:33 UTC+2:
"However, The PR itself is dependent onI have to commit that this sentence doesn't make much sense to me, maybe some context is missing.
"the match length.
MK schrieb am 6. Juni 2023 um 11:40:33 UTC+2:
"I also think match winning chance is an
"increasing function of match length for
"a certain PR difference.
If you throw a biased coin (lets say 0,52 head)
you don't agree that the probability of winning
betting on heads is a function of throwing the
coin only once or best of 3 or best of 100?
"However, The PR itself is dependent on
"the match length.
I have to commit that this sentence doesn't make
much sense to me, maybe some context is missing.
What might be meant is that the longer the match
is, the higher the probability is that the measured
PR is close to the real PR.
One medium error in a 1-pointer ruins your PR
whereas in a 25-pointer it hasn't much influence.
This is true (and rather trivial)
I'm to limited to draw from this the conclusion that
this shows a circular argumentation.
The original poster said that he doesn't know
his match equity table for longer matches.
I suspect that the effect of not knowing the
MET is very small in practice.
But there could be other factors, such as .....
This is just the point. If the formula was correct, theThen throwing a coin 2 times, 4 times 8, 16, 1026 etc. has alway come up with 0.5 heads? Then e.g. a 3-PR player to make an equivalent error in any subset of games. I recommend you to watch a UBC match with real time analysis. Or throw coin and count.
actual/estimated PRs would be accurate across all
match lengths.
Sure but the question is how did you determineF: Given A then B follows because of...
that the coin is 52% biased in the first place..?
MK schrieb am 7. Juni 2023 um 18:36:30 UTC+2:
This is just the point. If the formula was correct,
the actual/estimated PRs would be accurate
across all match lengths.
Then throwing a coin 2 times, 4 times 8, 16, 1026
etc. has alway come up with 0.5 heads?
Sure but the question is how did you determine
that the coin is 52% biased in the first place..?
F: Given A then B follows because of...
MK: how can you be sure that you have an A?
A typical MK answer.
doubling cube shortens matches
For example, you can run cubeful and cubeless matches between unequal
players to see if the winning probability does indeed increase more
slowly because of the cube.
Art Grater explains that Kaufmann created the
ELO for gamblegammon by modifying the ELO
for chess, (i.e. replacing D/400 by D times the
square root of N/2000). The problem goes back
to that. Since my first days on FIBS in 1997, I
always objected to it and called it a horse-fart
based on arbitrary constant. I'm glad to see it
being questioned so many years later.
It is easy to confirm, try it yourself.
On 6/7/2023 2:14 PM, MK wrote:
..... Since my first days on FIBS in 1997, I
always objected to it and called it a horse-fart
based on arbitrary constant. I'm glad to see it
being questioned so many years later.
I'm sure you raised doubts about it many years
before I did, but this is far from the first time
I've expressed doubts about it. See for example: https://www.bgonline.org/forums/webbbs_config.pl?read=182417
Axel Reichert <ma...@axel-reichert.de> writes:
I would think that the much increased
margin of the expert's win may safely
be called "cube skill".
But for sure I will be told here (not:
"learn") what I did wrong.
MK <mu...@compuplus.net> writes:
doubling cube shortens matches
Not that much, see
https://www.bkgm.com/rgb/rgb.cgi?view+712
... but it increases the number of skillful
decisions by at least 50 per cent,
so easily compensates for the smaller
number of checker moves.
For example, you can run cubeful and cubeless
matches between unequal players to see if the
winning probability does indeed increase more
slowly because of the cube.
I have not varied the match length, but done
something similar:
10 matches to 64 points, "Expert" versus "Casual
player" on GNU Backgammon. 1 batch of 10
matches cubeless, 1 batch of 10 matches cubeful.
The casual players on average won 10.2 points
in the cubeful matches, but 18.3 points in the
cubeless matches.
To me this seems like a cube-skillful expert
gardener cutting back the lucky branches
It is easy to confirm, try it yourself.
I really couln't understand his response to you and
you haven't said anything more to clearly indicate
your stance. Would you mind explaining now?
To expand on what I said in a previous response to
you in this thread: A single number may work well
enough for backgammon but not gamblegammon.
Another three batches, this time a money session, stopped when the
expert reached 1000 points.
1. The casual player got 321 points when played cubeless (no cube skill
involved). This is the base case.
Now two batches played with the cube:
2. When the casual checker play was combined with expert cube handling,
the casual player got 312 points, which is very similar to the base
case. As before, the size of the big win by the expert resulted only
from the checker play, since both players handled the cube expertly
(in contrast to "not at all" in case 1).
3. When the casual checker play was combined with cube handling set to
"Casual Player", the casual player reached only 103 points.
I would think that the much increased margin of the expert's win may
safely be called "cube skill".
On June 8, 2023 at 7:12:31 AM UTC-6, Axel Reichert wrote:
... but it increases the number of skillful
decisions by at least 50 per cent,
Who says?
On 6/8/2023 11:10 PM, MK wrote:
I really couln't understand his response to you and
you haven't said anything more to clearly indicate
your stance. Would you mind explaining now?
The reason I wasn't clear is that I didn't---and still
don't---have well-developed ideas of my own on
the subject.
To expand on what I said in a previous response to
you in this thread: A single number may work well
enough for backgammon but not gamblegammon.
Are you suggesting, perhaps, that if we were to
maintain separate Elo ratings for different match
lengths, then for cubeless backgammon, players'
Elo ratings would be pretty much the same for all
match lengths, but that with the cube, their Elo
ratings might not correlate very well?
.....
The difference between 0.05 noise and 0.1
noise is HUGE, even "Beginner" has only.....
MK <mu...@compuplus.net> writes:
On June 8, 2023 at 7:12:31 AM UTC-6, Axel Reichert wrote:
... but it increases the number of skillful
decisions by at least 50 per cent,
Who says?
Imagine the cube staying in the middle for
the whole game. Then obviously the number
of decisions has doubled
If the cube is turned .... then only the cube
owner has an additional cube decision,
which amounts to 50 per cent more decisions.
On June 12, 2023 at 5:51:50 AM UTC-6, Axel Reichert wrote:
.....
The difference between 0.05 noise and 0.1
noise is HUGE, even "Beginner" has only.....
I'm not trying to ignore you and would like to
discuss these kinds of subjects with you but
I honestly don't understand what exactly are
you trying to prove...?
I suspect it would be fairly easy to send random cube decisions but I
don't know if it is possible to query the bot for all legal moves and picl/send a random checker decision?
If you know how to do this,
why don't you have whatever levels of the "biased bot" play against
random cube and checker play?
MK <mu...@compuplus.net> writes:
why don't you have whatever levels of the
"biased bot" play against random cube
and checker play?
Because "I honestly don't understand what
exactly are you trying to prove".
Define precisely what checker/cube skill I
should pair against which. Maybe I can
detect a slight trace of meaning in the
setup and might do it.
Sysop: | Keyop |
---|---|
Location: | Huddersfield, West Yorkshire, UK |
Users: | 297 |
Nodes: | 16 (2 / 14) |
Uptime: | 15:23:06 |
Calls: | 6,667 |
Calls today: | 1 |
Files: | 12,216 |
Messages: | 5,336,691 |