This repository contains an analysis of GSGO tournaments data using SVM and NN. Two predictions are performed on the test data using aforementioned two classifiers. NN is configured to have 3 hidden layers with 2000 neurons each and ReLU activation. This is a rough implementation of classifiers without data augmentation. As one data augmentation technique, the order of the teams can be replaced, for example, from "team1 vs team2" to "team2 vs team1", and inverting "who_win" target value. This way the data can be increase twice for better generalization (not implemented due to time limitations). The input features are configured by merging the information of "players_feats.csv" and "train.csv" through "map_id" column. Some features, such as Players ID, are removed, since this info is redundant as other characteristics are given. Other features are normalized for faster training.
In order to rerun the program, it is necessary to install all relevent requirements listed in "requirements.txt" file using "pip install -r requirements.txt" command.