You can try this yourself without having to code.
I did it and played 10 games this way.
Here is how to do this:
1. Set GNU Backgammon to manual dice, "Expert"
play, "Expert" analysis, and start a new money session.
2. Roll a pair of real dice (different colours, e.g.
gnubg has the red one). Say, gnubg rolls a 4 and
you roll a 1. By now, gnubg has become "luck
master" and you will be the "luck servant".
closest roll to the target equity of 0.037 is 11. Move.
10. Repeat cycles from 2 to 6.
Yes, this was tedious, but kind of interesting. Some
impressions:
- As long as you do not blunder, the game stays very
close (the equity oscillates just a little around 0). A
pretty static tug-of-war.
- Doubles hence are rare (unless you have lost
considerable equity before)
- I won only one of the 10 games, as "luck master"
with the help of final lucky doublet in a pure race.
- It cuts down the branching factor considerably. For
example, after a 42 opening (played correctly) the
reply roll will always be 31.
After a big racing doublet by the master, expect the
same for the servant.
After a dance on the bar by the master, expect that
you will not be able to cover your home board blot
as a servant.
- Once I started to try to predict the rolls for the luck
servant (finally feeling like a dice paranoid instead
of working on my game) I was surprised quite often
by how bad doublets can be, or, put more generally,
how far off my estimations were. There is a lesson
to be learned by some here.
- There are highly artificial dice sequences: I saw 8
subsequent 21s for the servant in a race in order to
compensate for some previously accumulated luck
(that due to the position could not be compensated
adequately before).
- It can only be done with a computer.
- Considering the dice and luck paranoids out there,
the artificial sequences, the even more frustrating
outcome against a world class bot, I expect that
Gary Wong's official dice complaint form would
need a substantial update.
Not for me, but your mileage may vary.
On January 7, 2023 at 3:40:16 AM UTC-7, Axel Reichert wrote:
10. Repeat cycles from 2 to 6.
So far we are in agreement. Let's put a marker here.
On January 7, 2023 at 3:40:16 AM UTC-7, Axel Reichert wrote:
Have you saved your games? If so, can you paste them here in text
format?
- As long as you do not blunder, the game stays very close (the
equity oscillates just a little around 0). A pretty static
tug-of-war.
Why not blunder? To the contrary, my intention was to level the "luck
equity" but allow the "skill equity" swing and/or drift apart without limitations
- Doubles hence are rare (unless you have lost considerable equity
before)
I wouldn't have expected this but maybe because you eliminate
blunders.
what about cube actions in your games?
- It cuts down the branching factor considerably. For example, after
a 42 opening (played correctly) the reply roll will always be 31.
I'm not so sure about this. Replies to opening rolls can be memorized
but only assuming best play from both sides. Playing either roll
differently won't gain anything since it may not necessarily thwart
the opponent's subsequent correct play.
In any case, I foresee the branching factor going back to normal after
the 3rd roll.
- There are highly artificial dice sequences: I saw 8 subsequent 21s
for the servant in a race
I wonder if this will hold in the long run?
- Considering the dice and luck paranoids out there, the artificial
sequences, the even more frustrating outcome against a world class
bot, I expect that Gary Wong's official dice complaint form would
need a substantial update.
I don't understand this paragraph either.
If you mean you wouldn't enjoy playing this variant
gainst a bot, I can understand that but what about
using it in "bot vs. bot" and/or bot vs. "mutant bot"
experiments?
MK <mu...@compuplus.net> writes:
On January 7, 2023 at 3:40:16 AM UTC-7, Axel Reichert wrote:
Have you saved your games? If so, can you paste
them here in text format?
As you should know by now, I prefer if others get
their hands dirty as well. There is a lot to learn.
Why not blunder? To the contrary, my intention
was to level the "luck equity" but allow the "skill
equity" swing and/or drift apart without limitations
I understand this. But if I err/blunder (gnubg does
not), then the game does not stay close but rather
crawls slowly in direction of a double for gnubg.
- Doubles hence are rare (unless you have lost
considerable equity before)
I wouldn't have expected this but maybe because
you eliminate blunders.
I did not "eliminate" blunders, far from it. In fact I
was quite busy getting the "logistics" right and so
most likely paid less attention to the game itself.
For the record: I doubled not at all (but should have,
prior to the final 33 as "luck master", which won the
only game for me, an oversight by me). In all other
games gnubg doubled, I took 6 and passed 3. So
doubles are rare for the (non-expert) human.
what about cube actions in your games?
Well neither player adapted his strategy, but from
first principles I would argue as follows:
The luck master should be careful with doubling,
because there will not be many market losers.....
In any case, I foresee the branching factor going
back to normal after the 3rd roll.
No. The roll of the luck servant is fixed by the roll
of the luck master.
Backgammon is listed with a branching factor of
250 here:
I wonder if this will hold in the long run?
Anyway, this constraint was extremely annoying. The
game feels crippled, less lively. Try it yourself.
After all, it could reveal how weak their play is
(of course fueling further paranoia about the dice and
the luck calculations involved, which they most likely
do not even understand).
If you mean you wouldn't enjoy playing this variant
gainst a bot, I can understand that but what about
using it in "bot vs. bot" and/or bot vs. "mutant bot"
experiments?
No. I play agaist bots to prepare for play with humans.
Since I would not want to play this variant agaist
humans, this is "L'art pour l'art". Not interested.
As to your other post: I did things correctly.
All fine here.
I will not elaborate further,
My effort was not done in vain,
I will not play further games either.
Sysop: | Keyop |
---|---|
Location: | Huddersfield, West Yorkshire, UK |
Users: | 297 |
Nodes: | 16 (2 / 14) |
Uptime: | 13:44:20 |
Calls: | 6,667 |
Calls today: | 1 |
Files: | 12,214 |
Messages: | 5,336,513 |