The goal of this collaborative data science repository is to demonstrate how to transform and restructure data efficiently before running predictive models. The data used is > 20.000 competitive 1v1 ladder games from the open source game Battle for Wesnoth. The main focus will lie in explaining typical data preparation steps for modeling when dealing with a typical sparse dataset. Tidy data, visualization and modeling the data with different algorithms will accompany the project at various stages of its development. As a minium viable product, a predictive model to correctly categorize ladder game outcomes will be deployed and interpreted. The input data will likely be subset from the originally commited data file to prevent data leakage into the selected model, as the initial game file contains variables at game end, which strongly link to the victory condition. In summary, the proposed workflow will include the most important steps when dealing with data analysis: data pre-processing, visualization and interpretation of the predictive model. A documentation will include the rationale for selected steps in the process as the project progresses. Thank you for your interest in this collaborative project!
Load the data by typing the following commands in an R session
source('tidy_data.R')
load_wesnoth()