Skip to content

Latest commit

 

History

History
49 lines (42 loc) · 2.07 KB

README.md

File metadata and controls

49 lines (42 loc) · 2.07 KB

Music Recommendation Backend

Overview

This backend is part of a larger application, please see Music Recommendation Service for full details. The backend was built using Python and FastAPI, it is designed to act as a minimal shim between the vector database and the frontend. In order to keep deployment complexity to a minimum song data is included (./data) alongside application code. The backend assumes you have a Qdrant vector database pre-loaded with embeddings for the various songs. The vector database host is injected via the QDRANT_HOST variable.

Data Prep

Due to the size of the song dataset it hasn't been included as part of this repository. Please download the Ludwig Music Dataset from Kaggle to provide the needed song data. Once downloaded place the data into the ./data directory to match the structure below (yes, the mp3 directory is repeated).

├── metatdata_complete_music_data_set.json
└── mp3
    └── mp3
        ├── blues
        ├── classical
        ├── electronic
        ├── funk _ soul
        ├── hip hop
        ├── jazz
        ├── latin
        ├── pop
        ├── reggae
        └── rock

The following set of commands should do this:

cd $DOWNLOAD_LOCATION
unzip archive.zip
mv mp3/ $PARENT_DIR/music-recommendation-service/backend/data

Run

To run the application locally please ensure you have Python 3.x and pip installed then follow these steps:

# install the needed dependencies
pip install -r requirments.txt

# set this variable to point to your populated Qdrant database
export QDRANT_HOST=localhost

# start the API
uvicorn songs_api:app --port 8000

You should now be able to visit http://localhost:8000/songs or http://localhost:8000/songs?genre=electronic.

TODO

  • remove the song data and simply retrieve from Qdrant database
  • add troubleshooting section