About a year ago, I made a Spotify Recommendation project that I wasn't quite proud of, because it wouldn't work like a normal recommendation system. In this project, I used the movielens dataset to create a more realistic model. The model is hybrid ( it both uses content-based filtering and collaborative filtering ).
Then, I turned this model into an API using FastAPI and rendered it in a website. The results are quite satisfying to me.
If you want to test it and have it deployed on internet, here are the following steps:
- Install the repo :
git clone https://github.com/Brice-Vergnou/movie_recommendation.git
cd movie_recommendation
- Get the right libraries :
pip install -r requirements.txt- Deploy the web page locally thanks to the live server extension from Visual Studio
- Deploy the API locally :
uvicorn api:app --reload- Install ngrok
- Change your config file to :
authtoken: <your_token>
tunnels:
live:
proto: http
addr: 5500 # You can change the port to whatever you want, as long as it is the web port
api:
proto: http
addr: 8000 # Same thing with the API port- Start all your tunnels:
ngrok start --all
- Get a CORS proxy :
# From any folder
git clone https://github.com/Rob--W/cors-anywhere.git
cd cors-anywhere/
npm install
heroku create
git push heroku master
- Change the urls used in
script.jsin the first lines by your api server url and heroku proxy url