The work in this repository comprises a semester-long project in the course DATA 1030: Hands-on Data Science at Brown University, a core first-semester course in the Data Science Master’s program. I worked with the Automated Theorem Proving Dataset from the UCI Machine Learning Repository and applied the techniques learned in class throughout the semester, culminating in the final report which can be found in the report/ directory or by clicking here.
The work in this project was influenced and inspired by the dissertation by Dr. James P. Bridge at University of Cambridge. The link to the dataset on the UCI Machine Learning Repository website can be found here. The dataset was generated by Dr. Bridge’s work, and in his paper, suggestions for further work included studying alternative class labeling. I decided to build on Dr. Bridge’s idea by converting the time columns used to generate the target variable into a multi-label indicator target variable, which allowed multiple heuristics to tie for the fastest time.
The Jupyter notebooks found in the src/ directory explore creating the target variable, exploratory data analysis, and the experimentation with various machine learning algorithms on the dataset. Saved models for three of the algorithms can be found in the results/ directory, however, the file for random forest was too large to save in full, so only the best random forest model from my results is included.
Some figures I produced are saved in the figures/ directory (and several can be seen in the report with captions as well). All data files used in the project are available in the data/ directory.
To view any of my notebooks, you may install the conda environment used to generate all code in those files. Please run conda env create -n ml_atp -f ml_atp.yml from the main project directory.