Discogs Rec

Discogs Rec is a recommender system designed to enhance the Discogs digging experience by allowing users to input URLs of releases that they were interested in and receive the top N recommendations back, in order to aid in discovering new music that they might not have discovered on their own. It utilizes Spotify's Annoy library and is trained on data from the Discogs data dumps, as well as additional data such as user wants, haves, and pricing collected via webscraping. At the moment Discogs Rec only supports recommendations for electronic music listed on Discogs. This was a personal project I worked on in my spare time, so by no means is it perfect.

Repository Structure

• src/: Contains the source code for the creation of the Recommender System. This includes the preprocessing of the features used to build the model, generating the .ann file, and creating the mappings between release ids and artist/track titles.
• app/: Contains the code for deploying and serving the Recommender System as a web application.
• data/: Directory for the training data (instructions on how to obtain that below), annoy index, config, and mappings.

Setup

1. In order to succesfully generate the Annoy Index you will need the actual dataset, which you can download here. Once the parquet file has been downloaded, place it in the /data/training_data directory.

2. Clone the repository:

git clone https://github.com/justinpakzad/discogs-rec

Docker Configuration

The docker-compose.yaml file configures two main services: discogs_rec_model for pre processing the data, generating the Annoy index, and data mappings.discogs_rec_app for the API and Streamlit components.

Shared volumes are configured for directories like data/mappings, data/ann_files, and data/discogs_rec_dataset, so that each service has access to necessary files without duplication.

Please note, in order to run the discogs_rec container, you will need at least 8gb of RAM.

Docker Commands for Building and Running Services Locally

To build and run the discogs_rec_model and discogs_rec_app services with docker-compose, use the following commands:

docker compose up --build discogs_rec_model

Or to specify a subset of features to be used:

docker compose run --build discogs_rec_model python main.py --features avg_rating low median high countries styles

Once finished you can then run the fast_api services and get your recommendations using the following:

docker compose up discogs_rec_app --build

Streamlit App Example

Notes

Feel free to experiment with the number of features used to generate the Annoy index file. I've experimented with using slightly fewer features and found it resulted in more random but interesting recommendations. You can also adjust the weight of the feature matrices for varied results in weights_dict which resides in the preprocessing.py file. Lastly, tweaking the number of trees (n_trees) in the Annoy file can enhance recommendation accuracy at the cost of a larger .ann file and increased memory usage during queries, as found in main.py. If you are having issues running this locally due to a compute issue, consider lowering the number of trees. This is an ongoing project, I will be trying to add/update new features and fix any bugs that may appear along the way.