Authors: Zoe Hancox and Samuel D Relton
This repository contains the code used in the paper Temporal Graph-Based CNNs (TG-CNNs) for Online Course Dropout Prediction.
Architecture of the unbranched TG-CNN model used in this code:
You can run these models by installing the requirements
pip install -r requirements.txt
and following the steps below...
The Jupyter Notebook dataframe_save.ipynb file converts the mooc_actions.tsv MOOC data file into a dataframe .pkl file (either containg the last 100 actions targetid_and_scaled_time_last100.pkl, or all 505 actions targetid_and_scaled_time_all.pkl) with columns user ID, action ID and the time between the actions. These processing scripts are not required to run the model, but can be used to change the number of actions if desired. The labels_save.ipynb file converts the final dropout label from mooc_action_labels.tsv into the labels.npy file for easier input when training the model.
To train and test the proposed unbranched TG-CNN model (with γ=1) run
python models\mooc_without_variable_gamma.py
To train and test the proposed unbranched TG-CNN model (with a variable γ parameter) run
python models\mooc_with_gamma.py
To train and test the proposed branched TG-CNN model (with γ=1) run
python models\branch_mooc_without_variable_gamma.py
To train and test the proposed branched TG-CNN model (with a variable γ parameter) run
python models\branch_mooc_with_gamma.py
The TG-CNN models could not be cross-validated due to the computation requirements exceeding the time limit of the high performance computer used in this project. Future work will involve validation of this model in a more robust manner.
To train the baseline LSTM model with 5-fold cross-validation run
python models\baseline_LSTM.py
To train the baseline RNN model with 5-fold cross-validation run
python models\baseline_RNN.py
Torch version 1.7.0, tensorflow 2.4.1, numpy 1.19.2, pandas 1.2.4, scikit-learn 0.23.1 and cuda 10.2.89 were used for creating these algorithms (as given in the requirements.txt file). This work made use of the facilities of the N8 Centre of Excellence in Computationally Intensive Research (N8 CIR) provided and funded by the N8 research partnership and EPSRC (Grant No. EP/T022167/1). The Centre is co-ordinated by the Universities of Durham, Manchester and York. A V100 GPU was used with an IBM POWER9 processor.
Graph network of actions and the elapsed time between them put into a 3D tensor form example:
Hancox, Z., Relton, S.D. (2022). Temporal Graph-Based CNNs (TG-CNNs) for Online Course Dropout Prediction. In: Ceci, M., Flesca, S., Masciari, E., Manco, G., Raś, Z.W. (eds) Foundations of Intelligent Systems. ISMIS 2022. Lecture Notes in Computer Science(), vol 13515. Springer, Cham. https://doi.org/10.1007/978-3-031-16564-1_34
If you make use of this code, the TG-CNN algorithm or the paper please cite the following paper:
@InProceedings{10.1007/978-3-031-16564-1_34,
author="Hancox, Zoe
and Relton, Samuel D.",
editor="Ceci, Michelangelo
and Flesca, Sergio
and Masciari, Elio
and Manco, Giuseppe
and Ra{\'{s}}, Zbigniew W.",
title="Temporal Graph-Based CNNs (TG-CNNs) for Online Course Dropout Prediction",
booktitle="Foundations of Intelligent Systems",
year="2022",
publisher="Springer International Publishing",
address="Cham",
pages="357--367"
}