On December 4, 2021 at 7:20:18 PM UTC-7, Tim Chow wrote:
On 12/4/2021 8:49 PM, MK wrote:
It's not even necessary for the human player to be trying to
play like the bots or be nearly as good as the bots to know
what moves the bots would "calculate" as giving the most
equity (although it sure would help) because some positions
are simple enough even for a mediocre player to get right.
Your problem is that you don't play with bots enough
I play a lot but sporadically and not for other reasons than
what you are talking about here. So, true in thet sense.
to be able to predict separately what XG 3-ply, XG 4-ply,
XGR+, XGR++, and rollouts would do.
At whatever strength a human can play, including all of
its aspects as adjusting his strategy to his opponen'ts
strength (whether bots or humans), is what it is. There
is no need to measure it and put a number on even if it
can be done.
Now, with that, if the purpose here is to predict the bot's
play, assuming that it's useful to do so, it becomes more
and more difficult to do it as the bot's levels gets lower.
In other words, with human players' abilities to predict
being constant, it would be easiest for them to predict
bots closer to their own strengths ("or higher" but not
as much "or lower"). Predicticting that a lower human
or bot player will play poorly doesn't help in predicting
how they will make worse moves. However, predicting
that a higher human or bot player will play well may, in
fact, help in predicting how they will make better moves.
If you played with bots more often, you would be able to
say, "Ah! In this position, I predict that XG 3-ply and XGR+
would play such-and-such, but XG 4-ply and XGR++ would
instead play so-and-so."
I found your argument interesting and I am elaborating
on it beyond some simply reactive comments. Hopefully
you will bother to explain more also.
First a question: is this how you guys discuss positions?
I haven't read enough of them nor paid enough attention
to notice this. If you guys indeed do so, can you explain
what are the benefits?
As for myself, regardless of how often I would play with
bots, I would never try doing so because I don't see any
or at least enough benefit from it, especially since it won't
make me an overall better player and surely not against
opponents of unknown strength.
It's because you're too lazy to distinguish between how
these different settings play that you get confused, and
make wrong predictions.
None of it is true. I put way too much effort into BG to be
called lazy. I just don't see any benefit from distinguishing
between different bot settings. (I will explain more below).
The position I posted last week didn't involve my making a
prediction but just an observation about a bot playing wrong
even according to its own nature (for lack of a better word).
It's ironic how that position was similar to the one used in
the user XG manual in explaining how giving even one more
chance of hitting makes a move worse per the bot's equity
calculations.
I think you say "confused" instead of surprised by such an
unexpected error by XGR++ in a simple position.
Now, more about why I don't play against lower settings of
bots since and starting with Jellyfish.
At the time, I had understood (whether correctly or not) that
JF was faking to play at lower strengths by injecting various
amount of "random noise" into its decisions. I though that
this would be the perfect/easiest way for a bot to cheat and
explain it away as something normal/expected/necessary.
So, I always played against its level 7(?) expecially since I
was also challenged to beat the best.
With Gnubg, I tried many many times playing against its lower
strengths mostly for speed reasons as the Grandmaster, and
later 4-ply, wes just unbearably slow. While doing that, I also
had realized that I was actually doing worse against the lower
settings (this can be a whole different discussion and I had
posted about it many times in the past). When even some "PR
sacrificing" :) moves by Grandmaster and 4-ply were explained
away by bullshit arguments like mini-rollouts, etc. I decided to
never play against lower settings of Gnubg also.
I don't play against XG's lower levels for the very same reason.
But because of your post, I played quite a few "moneygammon"
against XGR+ just as a refresher experiment. Although I felt like
playing against it was "easier", I didn't do any better (at least in
the short run). If I noticed and questioned any of its decisions
as "suspicious", it would be so much easier for you guys to say
"Duh, nothing surprising from a weaker bot."
BTW: Even today, I still believe that certain versions/editions of
some bots cheat (even if "a la film noir" with me as investigator
who catches clues and uncovers things that the audience can't. :))
MK
--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)