Pitch Type Prediction Using RNN

Welcome! This is the repository for Group 23's project 🦍, Pitch Type Prediction Using RNN. Here, you'll find everything you need to reproduce our results and explore our approach to predicting MLB pitch types using recurrent neural networks.

---

Steps to Use the Notebook

1. Open the notebook.ipynb file in Google Colab.
2. Follow the instructions in the "1 Project Setup" section to:
   • Install necessary libraries
   • Import dependencies
   • Download the dataset for the project
3. Once the setup is complete, run the cells in the notebook sequentially from top to bottom to reproduce our results.

Customizing the Dataset

By default, our code downloads pitching data for Hunter Greene during the 2024 season. If you'd like to explore data for a different pitcher or time range, you can easily modify the configuration. For example, to download Shota Imanaga's pitching data for the 2024 season, use the following settings:

### CONFIG ###
input_last_name = "Imanaga"
input_first_name = "Shota"
input_start_date = '2024-04-01'
input_end_date = '2024-11-02'
output_root = 'input_all_data_2024.csv'
### CONFIG ###

Example Dataset

To get a sense of what the downloaded CSV files should look like, check out the Example Dataset folder. It contains sample data for Mason Miller and Shota Imanaga's 2024 pitching performances.

Need Help?

If you have any questions or run into issues, feel free to reach out via email: [email protected]

---

What Else is in the Repo

Here's a quick overview of other folders in this repository:

• Meeting Minutes: Includes all nine meeting minutes reports from the semester.
• Proposal: Contains our proposal video, peer reviews of other teams' proposals, and our responses to feedback.
• Video & Presentation: Includes our project video slides, the final video, questions from other teams, and slides used during the presentation Q&A session.

Project Report

The complete report, titled "RNN-Based Pitch Type Prediction: Insights into Sequential Patterns and Strategy Refinement", is available in this repository. It details the methodology, experiments, and results of this project. You can find it here: RNN_Based_Pitch_Type_Prediction_Report.pdf.

Highlights from the report include:

• Accuracy improvement from 55% (Bayesian baseline) to 69% using RNN models.
• Techniques for handling imbalanced pitch type distributions with a generalist-specialist voting mechanism.
• Model comparisons between Simple RNN, LSTM, and Attention-based LSTM.

For more details, please refer to the PDF.