TF Learn and Legendary Pokemon
What makes you the very best like no one ever was?
This time I wanted to used a more sophisticated model to try to predict whether a pokemon would be legendary or not.
If you are new to this series, I try to provide in depth analysis on the stats behind pokemon.
In this example, I am using Google’s TF Learn which is a high level version of tensorflow to try to predict whether or not a pokemon will be legendary or not given their HP, Sp. Atk, and Sp. Defense. I believe there is not many patterns behind what makes a pokemon legendary so I only used HP, Sp. Atk, and Sp. Def as the features to feed into the network.
After using 650 of the 800 pokemon as a training set, I built a neural network using a non linear activation function. This would usually be used for image classification but I though pokemon were complex biological creatures.
My network received an 89% accuracy rating based off of the 150 test pokemon.
I tried to specifically test the network with Solgaleo and Entei but to my despair, my network thought that at best were a 38% chance at being a legendary pokemon.
I then tried Kobe Bryant which is one of the most legendary basketball players ever and the network still thought he had a 27% chance at being a legendary pokemon. I guess we will see when he is put on the ballet for the NBA Hall of Fame.