README.md

NeuralRandUCB

RL project for the fall 2021 semester of the IFT-7201 (RL) course @ Université Laval taught by Audrey Durand.

This repo contains the code needed to carry out the experiments described in the project report. We have taken and adapted the following code: NeuralTS, RandUCB, Personalized News Article Recommendation.

In addition, we used a sample of the R6A - Yahoo front page today module user click log dataset to perform the experiment using news article recommendation. This can be obtained from the following link.

This file must then be included in the following directories: articles_recommendation > dataset > R6

Make sure you install the required libraries:

pip install -r requirements.txt

Please note that the experiments were carried out with an NVIDIA RTX 3080 graphics card, which requires the most up-to-date version of Pytorch (1.11). This version may not work on your machine, and you may need to install an earlier version.

Experiment 1: NeuralRandUCB's performance with different configurations

To generate the results:

python neuralrandUCB_vs_neuralTS.py

To generate the graphs:

python neuralrandUCB_vs_neuralTS_plot.py

Experiment 2: NeuralRandUCB's performance compared to other algorithms

To generate the results:

python performance.py
python stats_analysis.py

To generate the graphs:

python performance_plot.py
python stats_analysis_plot.py

Experiment 3: NeuralRandUCB's robustness to reward delay

To generate the results:

python rewards_delay.py

To generate the graphs

python rewards_delay_plot.py

Experiment 4: News article recommendation with NeuralRandUCB

To generate the results:

python articles_recommendation/main.py