This is a submission for the Recommender challenge by @Sirajology on Youtube.
For this challenge I choose to create a program that recommends artists to users from Last.fm, it takes real data from users and uses a recommendation algorithm to generate the results.
The original dataset file can be found here, but I took only the 100k first lines for this challenge. The reason for this is that the original file has 17,559,530 lines and weights 1.6G, so I did not want to upload the entire thing or code a downloader on the "fetch_lastfm" function because it took a lot of time to download.
The challenge required you to compare the results of 3 loss functions, but I didnt do it because I didnt know how to. ):
- numpy (http://www.numpy.org/)
- scipy (https://www.scipy.org/)
- lightfm (https://github.com/lyst/lightfm)
Just run
recommend_challage.py
It will ask you to select 3 users, by id (from 0 to 2033), press Enter after every question and it will give you the recommended artists for the users you selected.
Credits to Siraj