[gblake]

Originally Posted by Lil' Lavery

My general point about defense was that his formula didn't factor in defensive play properly to create an accurate simulation....

And I agreed... What I challenged readers to do is to show that using a more accurate simulation will cause Bill's suggested method to produce an end-of-tournament ranking in which the contestants’ estimated ranks (that are a consequence of their performance during the tournament) are farther from their correct ranks than are the estimated ranks that result from using the current FVC method.

In a non-linear, multi-dimensional problem space such as the one you (LL) portray fairly well in your message, ranking the entities in the space (i.e. the robots) according to a 1-dimensional metric (which is exactly what BOTH Bill's suggestion and last season's FVC (and , FRC, FLL, NFL, NBA, NCAA, FIFA, etc.) methods do) results in a compromise. If you accept that the BW suggestion and the current FVC method intrinsically suffer from this compromise, then you hopefully come back to asking two questions.

• Will a more sophisticated simulation cause the intriguing results of the Bill’s first simulation to evaporate, or will the increased realism be mostly irrelevant to the result (i.e. will the BW method still create end-of-tournament ranks that are more accurate than last season’s FVC method?)?

• Is there anything other than creating an accurate 1-dimensional ranking (so that the “top” 8 teams can become alliance leaders) that should be used to choose the match parings (i.e. Should the number of other teams encountered on the field be important, or should creating close contests (weak-vs-weak, strong-vs-strong) be important), or…?

The question is not whether Bill's simulation is 100% realistic. The question remains: “Will increased realism change the result? Will the rank assigned by the experimenter (which will in itself be a compromise in a non-linear, multi-dimensional problem) be reflected more accurately by the results of a BW tournament or by the results of a last-season FVC tournament?”

Originally Posted by Lil' Lavery

Even though you only have the same partner once, you still interact with the same three other teams three matches in a row. In a six match tournament (fairly large by FVC standards), you will interact with only six other teams out of the entire field during qualifications.

Let’s do the arithmetic:
ASSUMPTIONS:

• Assume 32 teams, 1 field and 6 hours

• Further assume paces of 1 match per 6 minutes for current FVC methods and 3 matches per 12 minutes in the BW method.

CALCULATIONS

• Last Season method

o Current FVC appearances on the field = 7.5 per team = {[(6 Hrs x 60 Min/Hr) / (1 match / 6 min)] x (4 teams/match)}} / (32 teams /tournament)
o If each match is filled with a new ally and new opponents, a team will see 22.5 other teams on the field in a total of 7.5 matches.

o Note that last season’s methods did not guarantee that each match was a combination of teams that had never seen each other on the field before.

• Other method being discussed

o BW appearances on the field = 3.75 tri_matches per team = {[(6 Hrs x 60 Min/Hr) / (1 tri_match / 12 min)] x (4 teams/match)} / (32 teams /tournament)
o If each match is filled with a trio of other teams, a team will see 11.25 other teams on the field in a total of 11.25 matches. More matches, more time on the field, more accurate selection of the “Top 8”, fewer other teams.

• The suggested (BW) method with two fields available

o Choose to run two fields in parallel to get the non-trivial benefit of seeing more teams, in-person, on the field, as both ally and opponent. Do this in addition to using the BW suggestion to get the non-trivial benefit of a substantially more accurate end-of qualification ranking.

o Now each team sees 22.5 other teams on the field (same as in the single field FVC method).

o Each team gets to appear in 6.5 tri-matches. This results in 19.5 actual contests on the field, instead of the 15 they get from a two-field version of the current FVC method.

Assuming that the simulation’s accuracy holds up when/if the nonlinear effects of defensive ability against different types of scoring approaches are added to it (see above and other discussions about whether this is necessary), I still see Bill’s suggestion being worth some serious consideration. It has attractive advantages and tolerable disadvantages.

Originally Posted by Lil' Lavery

The reaction to the FRC events of this past season suggest that the FIRST community would not accept this (look at the multiple threads about the scheduling algorithm for the 2007 FRC season), and in addition, those same threads suggest that the FIRST community favors random match generation (in terms of teams participating, they still value space between matches).

Analogies to last season’s FRC are weak at best.

In addition to once again pointing out that I was one of the people who disliked last FRC season’s match scheduling algorithm, and pointing out that the FRC algorithm FIRST desired apparently did not get fully implemented, let me make a few other points.

• The BW suggestion should not be implemented in a surprise.

• The amount that the BW suggestion has teams opposing or allied with repeat teams is not so much more than the last FVC season’s method did

o The first match of any tri-match is exactly like a match in the last season method.

o The second match is similar to the many matches under last season’s method in which you encounter past allies/opponents.

o The third of the three is definitely something that would be very unusual under last season’s method.

 However, if everyone involved knows why it is occurring, and has been shown how it produces more accurate rankings; and if everyone has been reminded that they are getting more time on-the-field as a side benefit; then I think that contestants will, in general, take the change in stride, even if they don’t wholeheartedly embrace it at first.

Originally Posted by Lil' Lavery

A similar argument could be used for the scheduling algorithm put in place for the 2007 FRC competition, which seemed to be designed to create more competitive matches (a good thing right?).

To steal a phrase from you, that was something of a “social experimenting” attempt. Also, it was based on the flawed assumption that team number is a useful enough indicator of team success on the field.

I’m not 100% sure what that fiasco was supposed to accomplish, but it was implemented incorrectly, was based on flawed assumptions, was a surprise to general population, and so far as I know, can not be justified by any public simulations of its consequences. In this thread we (all of us) are not making those same mistakes.

Originally Posted by Lil' Lavery

While his scheduling method doesn't base teams on individual performance, he also gave math showing that a more proper ranking would be based on individual points scored, which, as I said, would be next to impossible, and would discredit defensive efforts.

You are going to have to point this out to me. (with a link or a quote). I don’t recall seeing this in the results or in his conclusions/conjectures derived from those results.

Are you perhaps thinking about his descriptions of the experiment set-up in which the experimenter knows/assigns a scoring ability to each simulated robot so that the experiment can later tell us if the final rankings are a good estimate of those assigned scoring abilities (scoring-ability is the one-dimensional metric that is used to order/rank the one-dimensional simulation’s robots 100% “correctly” when the experiment is set up).

Blake