Running a Dummy Tournament : Auto population of data

[Deepak Jois]

Hi
In preparation for Bangkok, I wrote a couple of small scripts to
generate a large set of inputs (venues, universities, teams etc), and
to randomly generate results. This will help me do performance
testing, and also to generally run through a dummy tournament without
the hassle of manually inputting data.

I put up a wiki page describing the process here

http://tabbie.wikidot.com/dummy-tournament

The source code is available here

http://tabbie.svn.sourceforge.net/viewvc/tabbie/trunk/performance/

Feedback is appreciated.

Deepak

[Meir Maor]

This may be a good place, to mention some thoughts on simulating good
random results which would correctly simulate a real debate.

I figure we give each player a random score from with normal
distribution averaging 75 and and with a standard deviation of 5
cutting off values outside the 60-90 range.
in each round a new random number is drawn for each player
determining how well he did in this specific round, this should
average 0 and have a standard deviation of 3 cutting off values
outside the -9,+9 range.
Cutting off means pick a different random number.

If the selected programing language doesn't support drawing
normally distributed numbers, the simple method is using the box-Muller
transform. http://en.wikipedia.org/wiki/Box-Muller_transform

The base value for each debater is added to the new value for each round,
this is the players

The numbers for standard deviation and cut-off points are based on some work
I did in the past analyzing results from Israeli nationals, however I
was working
with a fairly small tournament.

The system proposed here has several flaws:
primarily it ignores correlation between the performance of players on
a given team,
this is not crucial because we could infact simply analyze the
performance of the entire team and ignore the fact that it is made of
individuals. Should we wish to take this into account both the base
value and the per round value should be broken up into two parts
the first drawn per team(identical for the players on the team) and
the second drawn
per player, I never analyzed the proper values to use for the approach however.
If someone has complete data with personnel scores on a round per
round break up
from a large tournament I would like to try and analyze it for
statistical patterns.
I am also interested in more i testing more theories on the matter among them:
* stronger players are more consistent then weaker players.
* a strong room makes you perform better.

Me.

[Klaas van Schelven]

Hi Meir, others,

I can follow you up to this point. However, my question would be: in what way would this more correct tournament data also help us? As developers? I assume developers are the target audience of such simulations in the first place - right?

Klaas

[Deepak Jois]

On 11/11/07, Klaas van Schelven <klaasvan...@gmail.com> wrote:
> Hi Meir, others,
>
> I can follow you up to this point. However, my question would be: in what
> way would this more correct tournament data also help us? As developers? I
> assume developers are the target audience of such simulations in the first
> place - right?
>
> Klaas
>

Just to clarify further. I intended these scripts to be meant more for
performance/reliability testing than to mirror a real debating
tournament scenario.

However, like Klaas I would like to ask how would it be useful to
simulate *correct* tournament data. It would be interesting for sure
(at least for me). But how would it be useful?

[Meir Maor]

The usefullness is two-fold:
* make better tab algorithms, this allows you to check how well does a
certain tab system
sort the participants, how well does it find the top n teams?
* The analasys itself(even with no code) may provide intersting results.

Me.

[Klaas van Schelven]

Actually, it may be usefull, but in a slightly different context: it could provide another check on inputting of data... especially in later rounds, but also with hindsight on previous rounds. This could be used to determine both "freak judges" (people who value debaters much higher/lower than would otherwise be expected) and incorrect data input.... but I guess the approach to take would be on a different level.

klaas

[Deepak Jois]

On 11/12/07, Ciarán Lawlor <ciaran....@gmail.com> wrote:
> Deepak, thereis also a facility whereby you can currently get access to
> dummy information when testing on Smoothtournament.com.
>
I did not know about that!

Klaas, could you elaborate please.

[Klaas van Schelven]

Hi,

Yes, but just for the first round. there is no option to generate random results on smoothtournament.

So the available data just gives you a (rather big, maybe too big) sample of some teams and locations. In fact, I wanted to add some other options as well, like "sample tournament after 5 rounds". Deepak - it would be great if you could generate some of these (say after round 1 - 3 - 5 and 9), export them as SQL. I can then include them on smoothtournament so that we can offer some more play data for adjudicator allocation.

Klaas