MusicML

Final project for scalable machine learning class. Based on Transformer machine learning model.

Setup

First create a virtual environment inside the root of your cloned repository.

For Windows

python -m venv .venv

Then, activate your virtual Python environment. This is showing the PowerShell activation script. There are other scripts for Bash and Windows Command Prompt.

.venv\Scripts\Activate.ps1

For Mac

Python 3 already ships virtualenv, but if it’s not installed in your environment for some reason, you can install it via the package for your operating systems, otherwise you can install from pip:

pip install virtualenv

You can create and activate a virtualenv by:

python3 -m venv venv
. venv/bin/activate

Next, install this repository's pip requirements.

pip install -r requirements.txt

Finally, install PyTorch. This is how to install it for Python 3.7 on Windows with CUDA support.

pip install torch===1.3.1 torchvision===0.4.2 -f https://download.pytorch.org/whl/torch_stable.html

Or isntall it using Anaconda on Mac

conda install pytorch===1.3.1 torchvision===0.4.2 -c pytorch

Scripts

The mha.py script demonstrates the multihead attention mechanism described in Vaswani et al. paper Attention Is All You Need.

Dataset

The project levages The MAESTRO Dataset for trainning purpose.