ayush1695 / eg_game_analysis

Objective:

• Given a dataset of Starcraft player performance data in ranked games. We want to develop a model to predict a player’s rank using the information provided in the dataset.

• Documenting decision making throughout the model building process (EDA, ETL, modeling, evaluation, etc).

Approach

Code

main.ipynb is the file which contains all the analysis

Archive Code

main_with_feat_eng in the archive_folder can also be checked where model was built using some additional features using additional feature engineering steps. Although the model performance degraded in comparison to the simpler model used in main.ipynb

Config

• config.ini file contains the path and model related parameters
• Currently load saved model is set to False so that the code runs successfully.

Saved Model

As the size of tuned model of hyperopt is more than 100 mb, the trained and tuned model.pkl was added to .gitignore

Requirements

Run pip install -r requirements.txt in terminal to install all the packages used.

Run Time

Approximate run time is around 1.5 hours on CPU

References

• Sklearn, lightgbm, SHAP, PyCaret, optuna and other python packages.
• Wikipedia on certain gaming players trait understanding