EPFL ML Recommender System 2019
Project 2: build our own recommender system, and test its performance
For this choice of project task, you are supposed to predict good recommendations, e.g. of movies to users. We have acquired ratings of 10000 users for 1000 different items (think of movies). All ratings are integer values between 1 and 5 stars. No additional information is available on the movies or users.
All information of the task and some baselines are provided in Exercise 10
Please see also detailed instructions on the course github.
- data_train.csv - the training set. Each entry consists of an ID of the form r3_c6 (meaning row 3 column 6) and the value between 1-5 stars, given by the user for this user/movie combination
- sampleSubmission.csv - a sample submission file in the correct format. You have to predict the star ratings of the matrix entries specified in this file. In this dummy submission example, a rating of 3 is predicted for all positions in question.
Your collaborative filtering algorithm is evaluated according to the prediction error, measured by root-mean-squared error (RMSE).
Each participant is allowed to make 5 submissions per day (i.e. up to 15 submissions per team per day). Failed submissions (e.g. wrong submission file format) do not count.