The Python scripts contained in this directory are for data preprocessing and LightGBM model training. To set up the Python environment it is recommended to create a new Conda virtual environment from the environment.yml file, like so:
conda env create -f environment.yml
This will create a new Python 3.7 virtual environment named ir_analysis with all required dependencies. This can be activated by running the following command:
conda activate ir_analysis
The following sections describe how to use the provided Python scripts to prepare a custom dataset to train your own LightGBM model.
CS:GO demo files sometimes contain malformed data, leading to errors occuring whilst parsing. To identify .tagged.json files which contain ticks with unexpected values for fields like aliveCT, roundTime etc., run the find_corrupt.py script. Simply invoke the script, passing the directory containing your .tagged.json files like so:
python find_corrupt.py /path/to/tagged/files/dir
The script will iterate through all .tagged.json files in the specified directory, and report any files which appear to have errors. These reported files can be removed from the directory, stopping them from being added to both the training and evaluation datasets.
python create_train_val_csv.py /path/to/tagged/files/dir
python train_lightgbm.py -t train.csv -v val.csv -n 250