Quantifying Pressure in Cricket for building Expected Runs model

Update: The recipe data project can be found here - https://github.com/Rit-ctrl/Recipe-Data-Analysis.

Sorry for the inconvenience

Agenda

• Utilise LSTM model to build pressure index

• Combine this with current state variables to predict runs scored

• Use other methods from literature

• Build models without previous state information to ascertain if pressure is useful for predicting runs scored

• IPL play-by-play data obtained from Cricsheet.org

• Train on IPL seasons 2008-2021, test on 2022-2023

• Compare Mean squared error for the models to evaluate

  

Code structure

• utils.py : Contains helper functions for reading the input files

  • process_df : reads csv files and processes them
  • get_data : gets training and test data
  • xR_Model : Pytorch model definition for proposed model

• 00_Read.ipynb : Notebook for exploring the data. Useful if you want to take a look at the data

• 01_PI_baseline.ipynb : Notebook for calculating MSE for baseline pressure index PI₃

• 02_Results.ipynb : Notebook for calculating MSE for proposed model, as well as Linear Regression and xGBoost Model

• DLR.csv contains the Duckworth-Lewis Resource table used for calculating the PI₃