The "Analysis of Nabla scoring system" thread

[Treeburst155]

Hmmm....excellent question. I think applying Nabla would be the way to go because Ralph's 90 may not be as great of an accomplishment as Harry's 90. Nabla is the answer. Great observation Ace Pilot!

[Holien]

Originally posted by Ace Pilot:

Suppose the Axis subsection has the following scores:

Fred:.....50...50...50

Ralph:...90...10...50

Harry:...10...50...90

Should they tie or is this a good reason to apply Nabla?

Edited to add: Obviously, I picked the scores to favor the Nabla scoring system, which could produce a tied result also. But my understanding from reading parts of the scoring manual is that Nabla was designed to favor the player that scored well consistently relative to others, as opposed to a player that happens to get a high score in one scenario that carries him despite so-so scores in other scenarios.

If you did points from 1 - 3 for each game it would score as follows: -

Fred:.....2...1.5...1.5 = 5

Ralph:...3...1...1.5 = 5.5

Harry:...1...1.5...3 = 5.5

This is based on tied players spliting the points between them. I.e. Tied second place you get 2 + 1 / 2 = 1.5pts.

If you look at the scores Ralph and Harry are in fact equal to one another. So how do you seperate them?

H

[Holien]

TB posted too quick it would appear that you need to try and assess which 90% is better?

However, this could all be down to them both having easy games and Fred being stuck with tough games each time?

You will never isolate that out.

So just by trying to assess the best 90% you will always at times get it wrong?

[Treeburst155]

So we have six player sections requiring only three scenarios. We have exclusive Axis and Allied "teams", from which Nabla scores will be calculated, with no mixing of Axis and Allied scores. I like it.

[Treeburst155]

Originally posted by Holien:

TB posted too quick it would appear that you need to try and assess which 90% is better?

However, this could all be down to them both having easy games and Fred being stuck with tough games each time?

You will never isolate that out.

So just by trying to assess the best 90% you will always at times get it wrong?

Nabla scores would be derived from ALL tourney participants on a given side. That result is then used to determine the subsection winners. If you're in an easy section or sub-section, you will do better, of course. However, Nabla with a flat curve will reduce the effect of outlier results. That is the key to dealing with the quality of the direct competition.

[Ace Pilot]

Originally posted by Holien:

If you look at the scores Ralph and Harry are in fact equal to one another. So how do you seperate them?

H

I think you are back to using Nabla, which is intended to score a player based on how well they did relative to the tournament's median score for that scenario. By tweaking the scoring curve, you can decide how much or how little you want to credit outlier scores.

I also think it would be a good idea to implement some sort of wild card qualifier for the finals, e.g., the winner of each section advances, but so do the top "n" scorers, where n is dependent on the size of the tourney.

[Treeburst155]

If you use an odd number of sections, you will get an odd number of Allied winners and an odd number of Axis winners. To even things up for the finals, the top second place score from each side could be admitted to the finals.

[Treeburst155]

Hmmm...brainfart in my last post.

Any even number of second place finishers could be admitted to the finals as long as they were equally split between the Axis and Allied sides. The number of sections does not matter except when it comes to the number of scenarios needed for the finals.

[walpurgis nacht]

I'm not sure I understand who's been wronged here to inspire this thread? I've heard my game with BigDog (Maleme) mentioned repeatedly with little mention of anyone else save a few. So what that that score gave me that many Nabla points . . . .who exactly was cheated out of a section victory by it? Even if any of the suggested improvements to the scoring system were put in place, I don't see how it would change anything, for any of the sections, in terms of the final results.

Even if BigDog were a new player, which he wasn't (in fact he was the best opponent I had to face in my section), a scenario that's really tough for one side should reward the successful underdog with interest if he can pull it off.

[Treeburst155]

Now, now Walpurgis. You aren't sure who has been wronged because nobody is claiming they were wronged. We're discussing how to improve scoring over the current ROW method. We're also discussing tourney formats in general.

As for large victories from a very weak side, I believe the general consensus is that the opponent has at least as much to do with it as does the skill of the victor.

No matter how ROW is scored, you easily won Group 4. You're the champ without question.

[Treeburst155]

As far as inspiration for this thread, I believe there were several, including myself, who thought their final Nabla score did not correctly reflect their overall performance. Not that we were cheated out of a section victory; but our score seemed much lower than we expected based on on our overall performance. For example, I beat the average clearly in three games. I lost one clearly, and got stomped by you. My final score looks like I played crappy in all my games.

This is the sort of thing that prompted this thread.

[walpurgis nacht]

Originally posted by Treeburst155:

For example, I beat the average clearly in three games. I lost one clearly, and got stomped by you. My final score looks like I played crappy in all my games.

This is the sort of thing that prompted this thread.

Yeah but you know, that's just lipstick. I'm only noting that none of the section winners would change based on any of the suggestions I've read here.

[Treeburst155]

I believe I stated earlier that I thought nothing radical would happen to the final rankings if scored with the original Nabla system. Your Nabla score would just have been much closer to the rest of us. IOW, my one truly crappy performance would not have dragged my score down so much that it looked like ALL my games were crappy.

EDIT: Only a small percentage will win. However, the rest like to see how they stack up. Some failed to see a correlation between their performance and their final score.

[ June 15, 2005, 03:11 PM: Message edited by: Treeburst155 ]

[Cpl Carrot]

TB: Just for curiousity sake, what format would you need the current ROW scores to be in to apply the Nabla Formula?

R

[BigDog944]

Originally posted by Treeburst155:

EDIT: Only a small percentage will win. However, the rest like to see how they stack up. Some failed to see a correlation between their performance and their final score.

Well said, TB. I had almost no expectation of winning this tournament when I entered. However, I did want some kind of guage as to my current level of ability. The question of this thread is, did I get that? The answer seems to be, yes, but with a large margin of error, which in my case seems to have been on the negative side.

Ultimately it would be great if a system can be found that reduces that margin of error.

[BigDog944]

Originally posted by Walpurgis Nacht:

Even if BigDog were a new player, which he wasn't (in fact he was the best opponent I had to face in my section)...

Thanks, WN!

[Treeburst155]

All I need is a text file; BUT, all the player names must be EXACTLY the same for each scenario with no spaces (case sensitive too).

The general format is this:

# Scenario_A

playerA 10 playerB 90

playerC 20 playerD 80

This text file would have to be manually typed from the raw results, doublechecking for the exact same spelling of names in all five instances. Also, I think I noticed some descrepancy in the names from scenario to scenario. Replacements perhaps?

[Kingfish]

Originally posted by Treeburst155:

All I need is a text file; BUT, all the player names must be EXACTLY the same for each scenario with no spaces (case sensitive too).

The general format is this:

# Scenario_A

playerA 10 playerB 90

playerC 20 playerD 80

This text file would have to be manually typed from the raw results, doublechecking for the exact same spelling of names in all five instances. Also, I think I noticed some descrepancy in the names from scenario to scenario. Replacements perhaps?

I can't help but thinking we are about to come full circle with regards to ROW, not that I mind

[Cpl Carrot]

I see. I could produce that from the database I think.

[Treeburst155]

Originally posted by Kingfish:

]I can't help but thinking we are about to come full circle with regards to ROW, not that I mind

No-no, not full circle. I like the way you have recruited volunteers to help run the various aspects of the ROW tourneys. I would be willing to volunteer as the schedule maker and scorer. The two tasks are closely related as the schedules produced by the Nabla Scheduling program are essentially the same as the scoring program input file. The hassles just discussed with Cpl Carrot can thus be easily avoided.

You run the sign-up phase. I generate the sections and schedules. You secure and disperse the scenarios and briefings. I collect the scores, nag players about the deadline, click through unfinished games, and finally produce the Nabla scores (and raw scores)for posting by Cpl Carrot. Your other volunteers continue with their present duties.

Any format changes, such as those discussed in this thread would be up to you. I'm not sure the Nabla Scheduling program is adaptable to format changes; however, this would be a minor issue I could deal with.

[Treeburst155]

Originally posted by Cpl Carrot:

I see. I could produce that from the database I think.

It would be a tedious task Cpl Carrot. 72 names reproduced 5 times each, spelled EXACTLY the same. It would look like this for five scenarios:

# Scenario_A

PlayerA 50 PlayerB 50

PlayerC 80 PlayerD 20

...and so forth

# Scenario_B

# Scenario_C

# Scenario_D

# Scenario_E

It is a rather large text file with NO mistakes allowed. If replacement names are listed for any scenario, they must be listed in all five scenarios.

EDIT: You could run each scenario individually through the program like I have done for Moltke. However, this would mean manually computing the average (final) Nabla score for each of the 72 players.

[ June 15, 2005, 07:55 PM: Message edited by: Treeburst155 ]

[Cpl Carrot]

If someone can explain to me what 'sign' means in formula on page 30 I could possibly do the calculations myself as part of the scoring calculations. Afterall I already have the medain and scores already entered.

[Treeburst155]

"Sign" means "insert negative or positive sign here". Off hand, I can't remember how one determines which sign to use.

I used the formula manually with a calculator before the DOS program was finished. With practice it became very easy.

EDIT: It's positive or negative deviation from the median!

[ June 15, 2005, 08:30 PM: Message edited by: Treeburst155 ]

[Cpl Carrot]

The deviation from the median is going to be positive or negative anyway so why include it? Seems to me you can just ingore it? :-/

What value of alpha do you recoomend?

[Treeburst155]

I will put together the large input file if tourney organizers wish. I can send the Nabla scores to Cpl Carrot or Kingfish by sometime tomorrow afternoon. Just say the word.

Treeburst155 out.