This repository contains the code for my numerical experiments on the gym data during my Master's Thesis.

This repository contains both convenience functions for operating on the data as well as runnable scripts that compute different sorts of statistics on the data. Before everything, it is advisable to create a new virtual environment to avoid cluttering your system's native Python distribution.

This repository contains a Pipfile for usage with Pipenv; for more information, see the Pipenv docs.

Installation on macOS is as easy as

brew install pipenv

provided you have Homebrew installed on your Mac. Afterwards, dependencies can be installed from the Pipfile by simply running

pipenv install

in the top-level directory. The virtual environment can be activated by the

pipenv shell

command.

On top of the Python code, this repository also contains some functions written in Cython to speed up computations on the data. Read about Cython usage here. The compilation and installation process is handled by Python's setuptools package. For installation, from the top level directory complete these steps while inside your virtual environment:

cd regularity_analysis
python setup.py build_ext --inplace

Some experiments on the dataset are located inside the bin folder. These are based on a config file architecture: Command line arguments for statistical functions and interesting hyperparameters for these functions can be specified inside these yaml files. Upon running these scripts, the yaml files are then envoked and parsed by the PyYaml module. Some sample configuration files are located inside the cfg directory, though you can pass your own to the scripts by supplying the --config_file flag.

Running a script takes one line. From the root of the repository, run

python -m bin.<script_name>

Notice that the file is run as a module, so no .py should be added at the end. For example, for running the changepoint_detection script file inside bin, you should type

python -m bin.changepoint_detection

from the root directory. If you are working on the repository yourself, a local installation could make this process easier, I did not figure out in time how to do that.

The code leverages mostly pandas and numpy/scipy, as the dataset is small enough to fit into memory. Some data preparation code can be found inside the Jupyter Notebooks, for those, just fire up a Jupyter Lab instance.