Variational Temporal IRT (VTIRT)

This repository contains the code used for the experiments in the paper "Variational Temporal IRT: Fast, Accurate, and Explainable Inference of Dynamic Learner Proficiency", published in EDM'23: 16th International Conference on Educational Data Mining

Installation

Go to the root of the repository and run the following commands:

conda create --name vtirt python=3.9
conda activate vtirt
pip install -e .
cd vtirt
pip install -r requirements.txt

These commands will create a conda environment named vtirt and install the required packages. All source code is contained in the directory named vtirt.

Running Experiments

This repository has code for running different types of inference experiments: exp/svi.py (Stochastic Variational Inference), exp/hmc.py (Hamiltonian Monte Carlo), exp/vem.py (Variational EM), and exp/tskirt.py. VTIRT and VIBO are trained and evaluated using exp/svi.py, which can be executed with the following command:

python svi.py [-h] [--device DEVICE] [--infer-only] [--valid-once]
              [--overwrite] [--resume-training] [--run-id RUN_ID]
              config_path

positional arguments:
  config_path

options:
  -h, --help         show this help message and exit
  --device DEVICE
  --infer-only
  --valid-once
  --overwrite
  --resume-training
  --run-id RUN_ID

config_path is the path to the config json file for running each experiment. They can be generated by calling

python configs/[vem,tskirt,vtirt]/gen_scripts.py

When an experiment is run, it stores all performance results along with model checkpoints - if there is if any - under the out/ directory.