Fantasy AI

Data dragon

Download: https://ddragon.leagueoflegends.com/cdn/dragontail-11.15.1.tgz

Available versions: https://ddragon.leagueoflegends.com/api/versions.json

Replace the url at https://ddragon.leagueoflegends.com/cdn/dragontail-9.3.1.tgz

Create datasets

1. Move batches of raw data and the extracted data dragon folder to a DATA_FOLDER of your choice.
2. Run aggregate.py DATA_FOLDER to get aggregated discrete time steps from the raw events. The script also adds extra features which can be mined from but are not directly in the raw data.

Win prediction dataset

3. Run create_win_dataset.py DATA_FOLDER to create a dataset for win prediction. This extracts features from the aggregated dataset for win prediction and saves them to a csv file. It creates a data/win_dataset.csv file.

Macro prediction dataset

3. Run create_macro_dataset.py DATA_FOLDER to create a dataset for macro prediction. It adds targets and spatial features to the aggregated game states and transforms the features so that they can easily be used with a neural network.
4. Run split_macro_dataset.py DATA_FOLDER which text files train.txt, valid.txt and test.txt with ids of games for each split.
5. Run macro_dataset_to_samples.py DATA_FOLDER which stores all the sequences of the given history size on the disk so that they can be loaded with a random access. The samples are split into train, test and valid folders as per the output of the previous step.

Train models

Win prediction

1. Run predict_win.py DATA_FOLDER which tests different models and saves their results in output/win_prediction/model_accuracies.csv.

Macro prediction

1. Run predict_macro.py DATA_FOLDER.