zevergreenz / model_fusion_for_personalized_learning

Model Fusion for Personalized Learning

This repository implements all experiments in the paper Model Fusion for Personalized Learning (submitted and under review at ICML 2021).

Requirements

• Python >= 3.6
• numpy, scikit-learn, pytorch, panda, matplotlib, d2l
• Run the following to install the requirements

pip install -r requirements.txt

Instructions for running experiment

• Set the parameters BASE and PROJECT_NAME in utilities/util.py to configure the path to save the models and results. Default is /tmp/personalized-learning.
• Turn on/off the model alignment by setting do_alignment to true/false
• Run the code as follows

PYTHONPATH=/path/to/the/code/directory python sine_experiment.py
PYTHONPATH=/path/to/the/code/directory python movielens_experiment.py

Sine Experiment

• The code for running the sine regression experiment is in experiments/sine_experiment.py.

Movie-Lens Experiment

• The code for running the Movie-Lens recommendation experiment is in experiments/movielens_experiment.py.

MNIST Experiment

• The code for running the MNIST classification experiment is in experiments/mnist_experiment.py.
• Before running the above experiment, you may want to create some pre-trained models using the script experiments/mnist.py.

Meta Model implementation

• The meta model implementation is in models/meta.py