Recommendation Systems

This repository is the implementation of the recommendation system for Homework 2 in the Information Retrieval Class (CSE272 Spring UCSC). All the code are written in python.

Prerequisites

Python 3

pytorch (pip install torch)

Guideline

Run data parser to generate training data and test data

Run command below:

python data_processor.py

This command will split the dataset into training data and test data and store the processed data in parsed_data/.

Run each algorithm to get MAE and RMSE evaluation:

In the repository, there are "user_CF.py","MF_train_eval.py" and "MF_train_eval_cuda.py". "user_CF.py" indicates user-based collaborative filtering, "MF_train_eval.py" indicates matrix factorization. "MF_train_eval_cuda.py" indicates matrix factorization implementation based on CUDA. One can simply run each algorithm via python command. For example, for user-based collaborative filtering, just run the command below to see evaluation of user_CF:

python user_CF.py

Plot Training dynamics

For matrix factorization method, one can also view the training dynamics of MAE and RMSE. Simply run:

CUDA_VISIBLE_DEVICES=0 nohup python MF_train_eval_cuda.py > results/MF_cuda.out 2>&1 &.

Then run:

python plot_results.py