DeepGameInference
Python program for training AI on video game data
Page last updated on 7/10/2021
I have been intrigued by neural networks for a very long time (I first learned about them when deepfakes were becoming popular) and with this project, I got to train my own neural network! Using PyTorch, a Python API for training neural networks, I wrote several Python scripts which will guess how much a user will like a video game they've never played before, based on the video game's properties. This works by requiring the user to compile a list of video games they've played and assign a "likability score" (a number from 0 to 10) which represents how much they enjoyed playing the game. The Python scripts will query more data for these games online (such as ESRB ratings, Metacritic scores, release dates, etc) and train a neural network on the data which maps these categories into the likability scores. Once the neural network is trained, it can be used to infer the likability scores of games which weren't part of the user's input game list. Currently, the neural network only considers the linear relationship between the properties and the likability score, but the program could easily be expanded or altered to feature a more complex neural network. The code is open source on GitHub and licensed under GPL 3, so anyone is free to fork the project and make improvements to it!