manymusic-mtg-jamendo

Streamlit app to create subsets of The MTG Jamendo dataset

Install

0. Clone this repo: git clone https://github.com/seunggookim/manymusic-mtg-jamendo.git

1. Get mtg-jamendo-dataset as a submodule: git submodule init && git submodule update

2. Create a virtual env: python3 -m venv venv

3. Activate a virtual env:

   • Mac/Linux: source venv/bin/activate
   • Windows: source venv/Scripts/activate

4. Install the Python dependencies: pip install -r requirements.txt

Note

An additional requirements file requirements-dev.txt should be used to run other functionalities apart from the Streamlit annotation app.

(not needed for annotation) Generation of the ManyMusic song pre-selection

1. Copy the data into data/. The required files are: mtg-jamendo-predictions-algos.pk, mtg-jamendo-predictions-av.pk, mtg-jamendo-predictions.tsv, and the timewise predictions/.

2. Start the app: streamlit run manymusic-viz.py

3. The Streamlit app will generate JSON file data/clean_tids.json with the candidate MTG Jamendo ids for the ManyMusic dataset. The resulting ids are randomly sampled from a pool of valid ids created with several filtering stages where the threshold can be updated by the user.

4. Run python clustering.py to generate a dictionary of tids sampled by applying clustering to the tracks belonging to the different genres.

5. Run python postprocess.py to generate a tsv combining several output jsons. Optionally, the resulting dataset can be split into equally sized chunks.

Annotation of the ManyMusic song pre-selection

1. Go to the cloned directory and activate the virtual environment (VENV):
   • Mac/Linux: source venv/bin/activate
   • Windows: source venv/Scripts/activate
2. Run the script: streamlit run manymusic-annotator.py