Saturday, August 08, 2009

Diamonds, Diamonds, Everywhere.....

For those following along in the message boards, IrishRef suggested a different algorithm for calculating blue diamonds. His idea was to throw the highest rated vote and lowest rated vote for each box out (those pesky "outliers") then average the rest and sort accordingly. It's an interesting algorithm, and not one I had considered before. I was intrigued--how would blue diamonds if I sorted it out that way?

And what about those people who think every vote should be counted "as is," no normalization of the votes allowed. How would that shake out? Sure, there would likely be a lot of overlap, but how much? Would one of these other algorithms provide better results? Ultimately, I'm not attached to any one particular algorithm. I'm more than happy to go with the one I think works best.

So I'm having an algorithm face-off. I have created seven, yes, count 'em SEVEN colors of diamonds: red, yellow, green, blue, purple, white, and brown. Each one uses a different algorithm to determine the 5% of boxes that will have that color.

I'm not going to tell you which algorithm goes to which color, or even what all the algorithms are. I will say, however, that one of them does use IrishRef's suggestion. Additionally, one of them is a "flat average"--it takes the votes as is and averages picking those with the highest averages. I'd like your opinions on which color you feel best represents the real "blue diamond letterboxes." I'd also like to point out the the color blue is NOT using the old algorithm--I've actually tweaked the old algorithm and given it a different color to disguise it a bit. =) None of the colors actually represent what the old algorithm used.

If one of the algorithms is a particular clear-cut favorite, I might update the code to use the new algorithm instead of the old one. =)

But please, be honest. Don't pick the color that gives your plants the most diamonds. Pick the one that you feel provides the most accurate results. These are supposed to be the best boxes out there--those that a visitor "can't miss" if they're passing through. Be honest with yourself, and select the algorithm you feel accomplishes this goal.

You can cast your vote in this post.

Red = funny
Yellow = interesting
green = educational
blue = agree
purple = disagree
white = huh?
brown = hug

I'm very curious to see how you all think the different algorithms stack up against each other. =) Also keep in mind, anyone who has opted out of the blue diamonds will not have ANY color on their boxes, so don't fault an algorithm for not putting a diamond on a boxer if you know they've opted out. The problem might not be the algorithm.

I also want to point out--all these colors are temporary. Eventually, I will be selecting ONE algorithm, and that's what'll be used for blue diamonds. The rest of the colors will go away.

Happy trails!

-- Ryan


Anonymous said...

I don't get how the = factors in with selecting a color.

Teresa said...

Oooo, look at all the diamonds! Seriously, it's fascinating to see how the different algorithms used effect which boxes get diamonds. I'm writing down notes on some of these boxes with various diamonds before they disappear!

Anonymous said...

Wow, those diamonds are all over the map. Hard to tell which algorithm is "better", since all the boxes that got diamonds have something going for them. The one thing that DOES seem to work: The boxes that I consider truly exceptional seem to all have garnered MULTIPLE diamonds, three or more.

-- Kirbert

Anonymous said...

I'm having trouble identifying the colors. Do you have a post where you can show the colors next to the word you use to describe them. Two different colors look like "pink"s to me and I don't see any brown????

Anonymous said...

Nevermind--I found the post with a clarification of orange/brown and red/pink. That helped.

Anonymous said...

I have found about 100 AQ listed letterboxes and placed 30. These are the boxes I know best, yet I feel I do not have enough info to choose the "best" diamond.

However, both the new BLUE and WHITE diamonds are decidedly NOT representative of boxes I would consider the best.

Hope that helps.

River Sol

Patrick P. said...

It seems to me that the truly exceptional boxes are the ones that get more than 1 diamond. Perhaps that is how they should be calculated. DO we really need another algorithm for figuring blue diamonds? I am not sure that the outliers pose much of a problem expecially with increases in the number of votes...the outliers become more and more insignificant.

It looks like all the truly exceptional boxes have been RED, YELLOW, or BROWN...singly or in combination. The very best boxes seem to have 2 or more of these colors

I think the factors that matter most in voting blue diamonds are:

1) the number of votes in the 4-5 range
2) the number of boxes that the voter has the more boxes a particular person has found, the more accurate their vote in theory
3) their voting history ... ie. the bell shape curvature of their votes

...perhaps a 1-5 scale doesn't quite capture the experience of some boxes. For instance some boxes are found at the end of incredible breathtaking hikes or have clever clues. Perhaps there should be a way to factor that in?

Anonymous said...

there is acutally good mathematical basis for using the outlier elimnation method.
You might also take those that have the smallest standard deviation after you average them,, that way there is better consensus around the ranking.

if there are 10,000 active boxes , you could pick the top 500 ,,highest average and smallests SD. I would also take out all retired, unknown and deleted boxes from the sampling... no one is ever going to look for them so how can it help?

Patrick P. said...

It looks like BROWN and YELLOW represent the better boxes

Autumn's Folly said...

Brown seems to represent the greatest number of what I consider to be awesome boxes, however - in looking again, there are some average boxes also getting the brown diamond.

I concur with others who suggested that the final result goes to a combination of algorhythms. :)