Post by WayoshiM on Oct 13, 2011 0:19:12 GMT -5
Hi guys,
The multi-game analysis feature of WOFTracker is "basically" operational - all the major stats of a single that it records have somehow been combined together. There is still a lot of refining to do to get relevant stats, though, and as mid-terms and a paper are fast-approaching for me, that will be awhile away. Also, the report for all episodes so far is over 1600 lines - I'm going to have to figure out how to split that up...
Anyways, one thing I'd like to refine early is how to analyze the BR to determine the best letters to choose. Last year on Sony boards, cheezyguyty used Hammer's site and the huge archive of the boards (I think about n = 1500) to determine BGHO. I know n = 18 so far is much lower compared to that, but take a look at this data:
The first stat, a "BarChart" of letters as is coded in the program, says that A has been called 10/18 times by players, and C 13/18 times, etc. (Note with just one WC BR use so far, 18*4 + 1 is in fact 73.) So CDMA is by far the most called letters for this season.
The next stat is a subset of my puzzle dissections - how much of each letter is in each puzzle. Each game has a "BarChartPartition" - a set of BarCharts mapped by some other variable (here, a round). The program takes the BarChart for the BR for each game and combines them into the one BarChart chart. Here we see that B is the best consonant, then D, then C/H, and finally O.
However, isn't it true that the frequency of the letter in a BR doesn't matter that much - it's about having at least 1? That's what the last stat does - it starts with an empty BarChart of letters, combs all those BR BarChart puzzleDissections and accounts for every non-RSTLNE non-dud with a +1, no matter what frequency, and 0 otherwise. Therefore, stat 3 is, loosely, a sub-BarChart of stat 2, the number of times a callable BR letter actually helps the contestant. The total (116) is honestly not that useful, but it's printed out for all BarCharts.
From this, we still get BCDO as the best calls, but it's close - G, H, P and Y all look good for a WC 4th consonant, and the vowels are surprisingly tight.
===============================
Getting C and D as top consonants is bothering me right now, but it might just be that n = 18 is just way too small for something like this, and that even though we're 4-14 so far, the puzzles have been set up for significantly better, meaning more Cs & Ds.
My question is: does anyone remember how cheezyguyty did his analysis? Did he somehow factor both puzzle dissections and strict +1/+0 into his work (he could have, since Hammer lists contestant letters) in some type of formula? And either way, does anyone else have any suggestions how to refine this analysis further?
The multi-game analysis feature of WOFTracker is "basically" operational - all the major stats of a single that it records have somehow been combined together. There is still a lot of refining to do to get relevant stats, though, and as mid-terms and a paper are fast-approaching for me, that will be awhile away. Also, the report for all episodes so far is over 1600 lines - I'm going to have to figure out how to split that up...
Anyways, one thing I'd like to refine early is how to analyze the BR to determine the best letters to choose. Last year on Sony boards, cheezyguyty used Hammer's site and the huge archive of the boards (I think about n = 1500) to determine BGHO. I know n = 18 so far is much lower compared to that, but take a look at this data:
BR analysis
Letters called in BR:
A: 10
B: 4
C: 13
D: 10
F: 2
G: 3
H: 5
I: 2
M: 11
O: 6
P: 5
V: 1
Y: 1
TOTAL: 73
Puzzle dissection for BR only:
A: 11
B: 12
C: 8
D: 9
E: 17
F: 5
G: 6
H: 8
I: 13
J: 2
K: 4
L: 4
M: 5
N: 7
O: 18
P: 6
Q: 1
R: 9
S: 6
T: 8
U: 13
V: 2
W: 3
Y: 6
Z: 1
TOTAL: 184
Number of BRs a non-RSTLNE letter was in:
A: 10
B: 11
C: 8
D: 8
F: 5
G: 6
H: 6
I: 11
J: 2
K: 4
M: 4
O: 12
P: 6
Q: 1
U: 11
V: 2
W: 2
X: 0
Y: 6
Z: 1
TOTAL: 116
Letters called in BR:
A: 10
B: 4
C: 13
D: 10
F: 2
G: 3
H: 5
I: 2
M: 11
O: 6
P: 5
V: 1
Y: 1
TOTAL: 73
Puzzle dissection for BR only:
A: 11
B: 12
C: 8
D: 9
E: 17
F: 5
G: 6
H: 8
I: 13
J: 2
K: 4
L: 4
M: 5
N: 7
O: 18
P: 6
Q: 1
R: 9
S: 6
T: 8
U: 13
V: 2
W: 3
Y: 6
Z: 1
TOTAL: 184
Number of BRs a non-RSTLNE letter was in:
A: 10
B: 11
C: 8
D: 8
F: 5
G: 6
H: 6
I: 11
J: 2
K: 4
M: 4
O: 12
P: 6
Q: 1
U: 11
V: 2
W: 2
X: 0
Y: 6
Z: 1
TOTAL: 116
The first stat, a "BarChart" of letters as is coded in the program, says that A has been called 10/18 times by players, and C 13/18 times, etc. (Note with just one WC BR use so far, 18*4 + 1 is in fact 73.) So CDMA is by far the most called letters for this season.
The next stat is a subset of my puzzle dissections - how much of each letter is in each puzzle. Each game has a "BarChartPartition" - a set of BarCharts mapped by some other variable (here, a round). The program takes the BarChart for the BR for each game and combines them into the one BarChart chart. Here we see that B is the best consonant, then D, then C/H, and finally O.
However, isn't it true that the frequency of the letter in a BR doesn't matter that much - it's about having at least 1? That's what the last stat does - it starts with an empty BarChart of letters, combs all those BR BarChart puzzleDissections and accounts for every non-RSTLNE non-dud with a +1, no matter what frequency, and 0 otherwise. Therefore, stat 3 is, loosely, a sub-BarChart of stat 2, the number of times a callable BR letter actually helps the contestant. The total (116) is honestly not that useful, but it's printed out for all BarCharts.
From this, we still get BCDO as the best calls, but it's close - G, H, P and Y all look good for a WC 4th consonant, and the vowels are surprisingly tight.
===============================
Getting C and D as top consonants is bothering me right now, but it might just be that n = 18 is just way too small for something like this, and that even though we're 4-14 so far, the puzzles have been set up for significantly better, meaning more Cs & Ds.
My question is: does anyone remember how cheezyguyty did his analysis? Did he somehow factor both puzzle dissections and strict +1/+0 into his work (he could have, since Hammer lists contestant letters) in some type of formula? And either way, does anyone else have any suggestions how to refine this analysis further?
