^ Imgur album of results (Note that it is multiple images, not the one you see on the thumbnail only)
After seeing the original class survey results I couldn’t help but shake the feeling that these pie charts… weren’t great to read. I’ve seen some complaints from colorblind people not understanding a thing but even though I can see colors it was still quite a struggle to read it all properly. Some colors were way too close!
Beesides, it was quite hard to tell apart any but the biggest slices when it came to size.
So I’ve set out to try and show this data in a legible bar graph!
How did I do this?
Since I do not have access to the actual results, all I had to work with was the pie chart. So I had to extrapolate the data from there.
Here’s how I did that:
First, I printed out the pie charts to fit on an A4 paper. Then I took the biggest slices for which the % was given, and converted that % to degrees out of 360 for the full circle.
Then, I manually measured all the other slices within half a degree precision
(This picture had some errors that i corrected later, the top slice was 35deg, for example)
Then I took the known numbers and inserted them in this Java program I’ve written.
For the uninitiated, here’s what it does:
-
It takes the degree values of the given % slices, and puts these in a list for later use (remember this list!)
-
It also keeps track of the following things:
An array of the manually measured degree values
An array of the %'s of things. Unknown values are marked as 0 for the time being.
- Now, it will go over the array of all the unknown values from left to right. With each of these it does:
Go over all the numbers in the list and do the following:
Take the ratio of the measured degree and the given degree in the list, multiply it with the known percentage.
(this gives us an estimate of what the % of this slice is)
Add up the result with previous results
When it’s done doing that it:
-
Takes the average of the results: This is the % value of that slice
-
Converts this % back to degrees AND adds this average to the list mentioned earlier, because it is the most accurate measure of the slice.
Now it repeats these steps for the next measured degree, and this time it has an extra value to make it’s estimate with.
This is so that the smaller slices with more room for error get as many ACCURATE measurements to find my average from as possible.
-
Once it has found all the % values for each slice, it prints them out and
-
adds them up to show me how accurate this was.
FYI, the %s of the 4 charts (in order) added up to: 102.16%, 99.8%, 97.8% and 99.5%
This means that the found results may bee up to ~2% off from the found value
(So for example 10% may actually be between 10.2 and 9.8%)
Revisualizing the data
Now that I’ve extrapolated the data from the pie chart as accurately as I could, I needed to visualize it in my preferred way.
To do this, I put the found data in a .csv file and got to work on it using the Pandas library in Python.
Next, I plotted the data using the aforementioned library to do so.
(something like:)
df_survey['Main'].plot(kind='bar')
to plot the results from the ‘What class do you consider your Main’ as a bar graph.
Now that I had bar graphs such as This one I needed to refine them for publishing!
So I took these images and edited them in MSpaint (Don’t judge me), Converting the indices 0 to 13 to the corresponding class name, and coloring each bar after weapon class.
But I didn’t stop there!
Bonus
Next, I added another column to my data table, the ’ Weapon Class’ where each class gets assigned to the uh… weapon class they’re in (Swords, Bows, Wands)
I repeated my visualization process again to show the last 4 images in the album, To show the general popularity of a class such as staves and melees. Turns out almost 40% of the player mains a melee! [T]-^^ppe ^^btw.
At the end of the imgur album there is a link to another 4 images that shows what these last 4 images look like when instead the mean value of these is taken (Because there are 3 melee but only 1 katana class).
These were not included in the original post because the % values on the right side of that graph make no sense, since the mean value was taken. It only shows ratios of one another.
But what about those inaccuracies?
¯\_(ツ)_/¯
Some slices in the pie chart were 1px in size. That made them both 1.5 degrees and I can’t get any more accurate than that, even though they probably did not actually tie in reality.
Additionally, the % values were given in 0.1% at smallest so that means they are accurate up until 1/500th of the actual value, while my degrees are only accurate up until 1/360th of the original, so some loss applies there.
This was immediately visible, because all found degrees added up landed up to a 9 degree range off from 360. (2.4% off).
My program helped to narrow down the error here and there but in the end they are based off my prone-to-0.5deg-error measurements, so they will never bee 100%.
How long did this take?
Getting the degree values for all 4 pie charts and making the program took about ~4hrs, plotting it and editing the images took another ~2hrs and making this post was about half an hour. This adds up to about 6 1/2 hours, over the course of 2 days.
s
