Words Above Replacement: A Computational Approach to Analyzing Baseball Writing

This repository contains all code and files for the submission of my capstone project, Words Above Replacement: A Computational Approach to Analyzing Baseball Writing, as part of the completion of the CLMS program at Brandeis University.

All code was run in PyCharm on a Framework Laptop 13 with the following specifications:

• Python version 3.12.4
• OS: Windows 11 Home version 25H2
• CPU: AMD Ryzen 5 7640U (3.50 GHz)
• GPU: AMD Radeon 760M (integrated)
• RAM: 16 GB

Repository

Project requirements are listed in requirements.txt.

The project's paper, which describes the project's background, methodology, and results, is contained in the words-above-replacement-paper.md file.

All files generated during training and analysis of the main experiment are located in the results directory.

An interactive visualization of the embeddings in the repository using the TensorBoard Projector framework can be found at https://ledibr.github.io/words-above-replacement-visualizer/. The repository for this GitHub Pages site can be found at https://github.com/ledibr/words-above-replacement-visualizer.

For more information on the project and specific details on the repository's contents, see the index in the docs folder.

Running the demo

The demo.py file contains a script that replicates the embedding training and results visualization pipeline, beginning with a pre-existing mention_corpus.cor corpus file.

To run the demo, first clone this repository to your local machine (or fork it and then clone that, or really whatever you want to do to achieve the same end goal). Once available, run the following command while in the root directory:

python -m demo

or alternately:

python demo.py

The demo produces a number of files in a new demo_files directory, which generally correspond to the experiment files in the results directory. For a description of these files, see the "Results" section in index.md.

The primary exception to this rule is the vis sub-directory, which contains files for use with the public TensorBoard Projector located at https://projector.tensorflow.org/. To use, select the "Load" button and upload the tensor file and (optionally) one of the two metadata files, per the site's directions. The files include:

• war_word2vec_demo.model: Saved KeyedVectors of selected vocab (player set + interest words).
• war_word2vec_demo_metadata.tsv: Basic tensor metadata containing only vocab labels.
• war_word2vec_demo_metadata_annotated.tsv: Expanded tensor metadata containing "word", "type", "count", "position", and "team" fields.
• war_word2vec_demo_tensor.tsv: Tensor file of selected vocab.
• word_counts_demo.csv: Counts of tensor vocab.

While the vis files do not have equivalents in the results directory, they are equivalent to the war_word2vec_focused files/"WAR Word2Vec Players + Target Words" data in the project visualizer mentioned in the Repository section above. This allows you to replicate the embedding projections (for the target embeddings) without relying on my dedicated demo repository.