Hello All,
I disagree with Ian about both the condensation of rollout
results and about the removal of the Mersenne Twister reference.
About repeated rollout settings: I personally do not find it
tedious to have to do a little bit of scrolling to get through
rollout reports. Having the settings listed for each roll leaves no
room for doubt as to what they are. If someone not familiar with
GnuBG rollouts reads the report or if it is read at a future time the
settings used will still be clearly understood.
About the removal of the Mersenne Twister reference which is
really two issues: 1. Ian seems to be suggesting (He can correct me
if I misunderstood him) that all of the random number generators
other than the Mersenne Twister RNG should be removed from GnuBG. I
think that since the different generators are already in place they
should be left alone. Having a selection to choose from cannot be a
bad thing. Sure, it seems that almost all rollouts seem to be run
using the Mersenne Twister RNG. The reason most rollouts with GnuBG
use the Mersenne Twister generator could be because that is the
default choice and no one changes it. If the present code does not
require a high level of maintenance if should be kept. The most
recent change appears to be three months ago to the MD5 generator and
longer for the others. 2. Obviously if there is a choice of
generators, rollouts must state which of them is being used. However,
even if the Mersenne Twister generator is the only one available it
would still be a mistake to remove the reference to it in rollout
reports. Removing the reference includes removing the statement of
the seed used because only stating the seed would lead to confusion
as to what the RNG was. If I understand correctly, the seed is stated
so that if there is doubt about the results the rollout can be re-
created using the same seed. You have to know what the generator is
for the seed to be meaningful. Even for the sake of completeness it
is important to state all of the rollout parameters.
That's just my two cents on the matter. Take care,
Myshkin
On Wed, Jun 10, 2009 at 6:39 PM, Ian Shaw<[email protected]>
wrote:
I would like to see a more space-efficient report of rollout results.
Currently, when you see web pages with gnubg rollout results included
(typically gamnonu or bgonline), an awful lot of space is taken up
with
repeated rollout settings. This entails a lot of scrolling and
makes it
difficult to read the actual results.
I suggest only printing a parameter when it differs from the next
trial
DOWN the list, or is the last one on the list.
---------cut ------------
In fact, I wouldn't be sad to see the Mersenne Twister reference
disappear. I've never seen anyone post a rollout using anything
else. In
the early days of gnubg, having lots of RNG's was a good idea in the
interest of transparency. Nowadays, no serious user (one likely to
read
a rollout) doubts that gnubg is honest, so there is not really a
need to
provide the choice.
_______________________________________________
Bug-gnubg mailing list
[email protected]
http://lists.gnu.org/mailman/listinfo/bug-gnubg
