Machine Learning Project 2: Recommender Systems

The goal of this project is to apply machine learning techniques to get a very precise discrete recommender system.

Getting started

In order to run our code, all you need to do is to type in the console at the location of run.py:

python run.py 

Note that it can be quite long to run (up to 30 mins) because it has to train multiple algorithms. We have put some print function to help to see what's going on while running the file.

Dependencies

The external libraries that we use are surprise, time, pandas, numpy, scipy, csv, tensorflow, keras and sklearn. All of these libraries are very easy to install, and you can use the following commands if you miss some of them:

pip install numpy
pip install scipy
pip install pandas
pip install scikit-surprise
pip install -U scikit-learn
pip install tensorflow

Required files

In order to be executed, the run.py file needs the following files in the same folder:

data_formatting.py
implementations.py
neural_network_regression.py
discrete_surprise.py

All those files should be already included in the code.zip file at the right location. The data csv files (data_train.csv and sampleSumbmission.csv) must be located in the folder csv (so their path from run.py is csv/data_train.csv and csv/sampleSubmission.csv). The output submission.csv, and the preprocessed data_clean.csv files will be located in csv as well.

Exact reproducibility

In order for our code to reproduce exactly our best CrowdAi submission, we set all the random_state parameters of both surprise and sklearn to the value 2018. Our code should be deterministic.

neural_network_regression.py

This file is an example of how we used neural networks for our regression. It requires tensorflow and the same dependencies as for run.py It trains a model on a subset of our training set for regression and outputs the predictions for the remaining test set in a csv file.