Hipster Game Research Paper

679 Words3 Pages
--- layout: post title: The Hipster Game, or, a Very Serious Introduction to Core Concepts in Supervised Learning --- ![Can we identify a Hipster based solely on appearance?](https://github.com/GarrettHoffman/garretthoffman.github.io/tree/master/assets/hipstergame_banner.jpg "The Hipster Game, or, a Very Serious Introduction to Core Concepts in Supervised Learning") A quick [Wikipedia](https://en.wikipedia.org/wiki/Hipster_(contemporary_subculture) "Hipster (contemporary subculture)") search for "Hipster" describes a person beloning to "The subculture described as a "mutating, trans-Atlantic melting pot of styles, tastes and behavior" and that is broadly associated with indie and alternative music, a varied non-mainstream fashion sensibility…show more content…
Different groups took different approaches, some constructing a + or - points based system for Hipster and Non-Hipster qualities (such as clothes/style, facial hair, accessories, general "Hipsterness", etc.), respectively and others using a classification technique to first identify the definite Hipsters and Non-Hipsters and then drill down further on the tweeners. Our group opted to construct a set of rules that systematically rules out the Non-Hipsters and classifying the leftovers as Hipsters. The algorithm centered around style and deameanor and captured all of our known Hipsters and one known Non-Hipster (A6 above) which we decided not to add an additional rule to exclude (more on this…show more content…
![True Training Set Hipsters](https://github.com/GarrettHoffman/garretthoffman.github.io/tree/master/assets/hipstergame_trainset.jpg "True Training Set Hipsters") Our model classified 2 Hipsters and 10 Non-Hipsters correctly but misclassified 1 Non-Hipster as a Hipster (false positive) and 2 Hipsters as Non-Hipsters (false negatives). The algorithm predicted hipsters with 80% success, which appears pretty good at face value and slightly outperforms the most basic model of simply classify everyone as Non-Hipster (73% success). On the other end of the spectrum, if we added a 5th rule to our algorithm (e.g. has forearm tattoo) to exclude the known Non-Hipster while training the model we would have miss-classified another Hipster in our test set. This demonstrates examples of underfitting (simply calling everyone a Non-Hipster) vs. Overfitting (making a model that is too specific to our training set). ![Class Results](https://github.com/GarrettHoffman/garretthoffman.github.io/tree/master/assets/hipstergame_trainset.jpg "Class

More about Hipster Game Research Paper

Open Document