lycier, on 2016-April-23, 05:26, said:
Now the problem comes out, for opening 2♦ and passing bid, which is a Bug?
First, this example seems quite different, since both opening 2D weak on West and Pass are completely reasonable bids.
More a matter of personal style and "taste", and neither would classify as a "bug" to me.
But are you saying that Gib will sometimes randomize between which option to choose?
I have looked through many hand-travellers after playing robot-tournaments to see what happened at other tables,
and I am under the impression that, as long as different humans act exactly the same on a specific deal,
the robots will always take exactly the same actions, too.
Only after
humans take different actions
(might be only small differences like one human playing the 3 from 953 in a suit, while another human chooses the 5)
may the robots diverge in their responses.
This is a very good approach, of course, to achieve maximum fairness in the robot tournaments.
In programming, the obvious way to achieve this feature
(i.e. identical human actions at different tables is guaranteed to generate identical robot responses,
even though Gib uses monte-carlo random sampling in bidding an play)
is for each deal to
initialize the random-generator with exactly the same numeric seed at all tables.
However, I have mostly looked at pure robot-tournaments (always 1 human + 3 robots at every table),
so it might be that BBO for some reason
is not using the same approach when a robot is playing in a competition
with a mix of human-human and human-robot pairs, like your example shows.
I dont see any technical reason, though, why the same approach could not be used in such tournaments, too,
as long as the robots are playing the hands
from the start of the deal.
Only when a robot is substituted into the middle of deal (because a human left the table or was "timed out") it is generally not possible, of course.