Let me predict how you vote!

Posted by (twitter: @LiamLimeGames)
April 26th, 2015 3:38 pm

Well… not literally!

A few days ago I made a tool which analyses your votes. Shows you all sorts of things about how you rate jam games vs. how you rate compo games. Shows how many of a particular star rating you give for each category, etc.

Now I added one more thing! Machine learning! Now there is some code, which will attempt to predict your votes. Basically you feed it information about your votes – same as you did before, but now it tries to make a function out of those. Then it compares what it predicts with the actual votes you cast and checks how well it managed to predict your votes. The better it managed to do so, the more related these categories are. So, say you wanted to see how much the innovation votes influenced the overall votes you cast, you can do so by looking at the middle graph here:

In this case – quite a lot. Basically the lower the Linear regression error value is, the more related the two categories are.

I can only test this on my own votes, so go check it out for yourself. Basically a low number is bad and a high number is good. Unless you interpret it differently 😀

Information on the other features can be found in the post: Correlations??.

Get stats about your votes here

Disclaimers, you may skip this part: 

Note: All the processing is done in your browser, so don’t worry, the super sekrit voting data never leaves your browser. It’s all JavaScript!

Linear regression error calculated based on absolute distance, no modifications made to adjust for overfitting, model testing done on 100% of the training data. Should however be fine, as these models won’t be used to predict, just to analyse existing data.  If you’d like to analyse data yourself, open the console (F12) and enter ‘thedata’ and press enter. This variable contains all the data, the variable ‘linreg’ contains all the linear regression data.
Want it to provide more data? Let me know in a comment. Found a bug? Let me know in a comment! :D The data that is copied however does not include comments, so I can’t provide graphs that relate to comments. Sorry…
If the instructions on how to copy data aren’t clear, or if they don’t work properly in a browser or operating system, please let me know. I tested it on Chrome 42 on Windows.
Please don’t look at the JavaScript code too-too hard, as it’s a very ugly botch-job. It’s not code that is supposed to continue to work for years, just a quick and dirty implementation. If you would however like to expand on this, you can use it at will.

Get your stats! 😀

Leave a Reply

You must be logged in to post a comment.

[cache: storing page]