CIL-Collaborative-Filtering

Repo for the CIL group project on Recommender Systems and Collaborative FIltering using SVD

Updates after 3rd meeting on 7-May

1. Wrap up the finished functions into a class

2. cleaned up the workspace

Updates after 2nd meeting on 30th-Apr

1. Wrap up the .csv file processing functions into helper.py, which contains two methods:

• csv_parse(): read in the original csv file and parse the row/col index, then save a new csv with row/col ids for later use.
• write_submission(): given a matrix with predicted results, the function write out a csv file named "submission.csv" for submission

2. Optimised the creation of baseline solution

• Calculating the average of the non-zero elements in each colomn of matrix A is now handled in a vectorized way, which is more efficient.

3. SGD simple version done. Convergence at ~ RMS=0.6, score=1.2

• The SGD algorithm in Step-3 should be optimised for speed and covergence rate

Structure of the project

1. svd_approach.py

   • contains the baseline implementation of the recommender system using simple SVD

2. data_handler.py

   • contains all functions that parse the data, including loading and processing .csv files

3. paths.py

   • contains some paths and constants that are used throughout