Automatically Generated Piano Fingering using Q-Learning

This repository contains the code for automatically generating piano fingerings using a reinforcement learning agent that uses Q-Learning.

About the code

This code allows the user to input a number of music sheets in the MusicXML file format, which will be trained using the epsilon-greedy Q-Learning algorithm. The output of the program will be a pandas dataframe with each of the notes from the right hand, along with their predicted optimal fingering. A few MusicXML sheets have been provided and are stored in the data/train folder. To test the program, simply follow the "How to run" subsection of this README. Q-Learning is not good for generalization and this algorithm works best when the sheets in the test set are part of the training set. This ensures that the reinforcement learning agent has already seen the note combinations and the Q-Values are not empty (which results in standard finger 1).

Dependencies

The required packages to run this code can be found in the requirements.txt file. To run this file, execute the following code block:

$ pip install -r requirements.txt 

Alternatively, you can install the required packages manually like this:

$ pip install <package>

How to run

• Clone the repository
• Place all MusicXML .xml files that you want the reinforcement learning agent to train on in the data/train folder. Place all the MusicXML .xml files that you want the reinforcement learning agent to label in the data/test folder.
• Run $ python src/main.py
• See result