Clash Royale Battle Predictor

Neural network trained to predict the outcome of Clash Royale battles.

Details

• Takes in each player's deck and trophy counts. Returns the predicted outcome of their battle.
• Includes a data grabber /crbp/data_grabber.py to compile massive lists of battles (training data) through the Clash Royale API
• Scraped dataset has been uploaded: Visit Kaggle page here.

Predicting

1. Install libraries specified in requirements.txt
2. Run the file predictor.py
3. Use final for the model name. Models are stored as folders inside /models, the final model comes with the code.
4. Enter deck data and trophy counts. Case does not matter, but you must use the real card names e.g. "The Log" instead of "Log".
5. The resulting prediction has an "advantage" value which ranges from 0 to 1. It describes just how much stronger the winning player will be.

Contents

• predictor.py takes a trained model and runs user-given inputs through it.
• main.py is the full training process of the model. This file also allows saving of models. Models are saved as individual folders in the /models folder.
• main.ipynb is the full training process, but with detailed anotations.
• data_grabber.py collects training data by iterating through player battle logs in a list of Clash Royale clans. Output is saved at data/data_ord.csv, and data/cardlist.csv provides a key to translate card numbers into their names.