By ECE 229 Group 4: Arth D., Chin L., John O., Moira F., Tawaana H., Zizhan C.
This dashboard is developed by a group of music-lover graduate students as the final project for ECE 229: Computational Data Science & Product Development at UCSD.
If you're interested in learning more about your music listening preferences, this dashboard is for you. By signing in with your Spotify Credentials, you'll be able to see analysis and visualization of your music tastes, such as TSNE spatial clustering of all tracks, genre breakdown, top artists & top tracks, recommended tracks for you and more.
We hope Mus-X brings our users a pleasant visual & interactive experience, and help understand one's music preferences better.
Requires python 3.7+
Some main third-party modules:
- dash 1.20.0
- Flask 1.1.2
- pandas 1.1.4
- scikit-learn 0.23.2
- scikit-surprise 1.1.1
- plotly 4.13.0
- spotipy 2.18.0
Clone the repository using
git clone https://github.com/ArthDh/ECE229.git
Create a python virtual environment
python -m venv envActivate the environment
source env/bin/activate
Install dependencies
pip install requirements.txt
Deactivate when done making changes
deactivate
-Create a spotify developer account (its free) https://developer.spotify.com/dashboard/login
From https://developer.spotify.com/dashboard/applications :
- Select "create an app"
- Choose an app name and description, eg. "test", "test"
- Select "edit settings"
- Set redirct url, eg. http://127.0.0.1:5000
- From the application page select show "client secret" either (1) create a .env file ad save in under the 'app', 'util' and 'test' folders.
- .env will appear as follows:
SPOTIPY_CLIENT_ID='YOUR_CLIENT_ID'
SPOTIPY_CLIENT_SECRET='YOUR_CLIENT_SECRET'
SPOTIPY_REDIRECT_URI='YOUR_REDIRECT_URL'
or (2) in the python environment (linux only):
export SPOTIPY_CLIENT_ID='YOUR_CLIENT_ID'
export SPOTIPY_CLIENT_SECRET='YOUR_CLIENT_SECRET'
export SPOTIPY_REDIRECT_URI='YOUR_REDIRECT_URL'
To run this webapp on your machine, go to the directory where this repository sits, and type in your terminal:
python dashapp.pyThis will launch a server locally on your machine. Clicking on the url shown will bring you to Mus-X home page in your web browser. Then you can sign in with your Spotify credentials to start the jorney of understanding your music tastes!
Note: We have also deployed this web app on AWS for the period of this course, but the site will be down when our credits run out.
Note 2: Initial run of the app may not show the graphs as the CSVs are being built in the background. You should see the progress in the console. You should have 8 CSVs under .csv_caches/'YOUR_USER_ID' when the process is completed. You may have to restart the app and sign-in again to view the graphs.
Note 3: The recommendation system model are a couple of big file which can be downloaded from here: Model Link and need to be placed under app/assets/rec-files for the recommendation engine to work.
There are a several main milestones that defined this project.
We used Flask as the web framework, and integrated Dash by Plotly to create interactive data visualization plots that ties to modern UI elements. The web application is deployed on AWS EC2 Scaling Group, with S3 used for large files storage and Dynamo DB for session cache. Data analysis, processing and integration are coded in Python. The Spotify Python API and OAuth Security handles the Spotify user log-in credentials.
The plots in this dashcoard are coded using Plotly. We wrote callback functions to allow a more interactive data exploration experience for the user.
We used TSNE(T-Distributed Stochastic Neighbor Embedding) to cluster and visualize tracks in a users playlists based on the 13 audio features of each track. The user can select playlists of interest in the dropdown list to visualize the similarity of songs.
We built a user-item collaborative filtering recommender to generate personalized song recommendations for users. The model applies SVD (Probablistic Matrix Factorization) algorithm to learn user preferences of songs from external training dataset, and predicts scores of songs during inference. The user can export the recommended tracks playlist to Spotify by a simple button-click.
Documentation is created via Sphinx. From the section of the dashboard titled "check out our documentation" select the made "made with" to be redirected to our documentation page
Coverage report is generated from the coverage python package
chromedriver needs to be in the path variable, it can be added using:
export PATH=$PATH:/path/to/chromedriver
From the ECE229 (root) directory, run the following commands:
coverage run --source=app -m pytest
coverage report -m