The code for the EDM 2023 paper - A Conceptual Model for End-to-End Causal Discovery in Knowledge Tracing
The required environment is as bellow:
- Unix-based system
- Python 3.9+
- PyTorch 1.12.1
- Numpy 1.23.1
- Wandb 0.13.4
- Pandas 1.5.0
- SciPy 1.9.1
- HuggingFace (Transformers) 4.18.0
/data/ - Repository containing the dataset
/serialized_torch/ - Intermediate data from the prepocessed dataset
predict_graph.py - Training Causal KT
construct_solution.py - Creating adjacency matrix for submission
/submissions/final - Contains the model and the adjacency matrix for submission on the private leaderboard.
First install all dependecies of the projects below is the pip methodology. (Similiar approaches exist for conda or other package management systems).
pip3 install -r requirements.txt
Here is an example how to train the Causal GRU Model with a learning_rate of 0.01.
python3 predict_graph.py -L 1e-3
For a list of all possible hyperparameters see:
python3 predict_graph.py -h
The output of this is a model .pt file which contains a learned P and L matrix. By default, this is saved into the saved_models directory.
To construct the construct ordering adjacency matrix from this model we use construct_solution.py. Here is an example how to create a submission file. You do not need to include .pt.
python3 construct_solution.py -f <Your Model File>
The output of this script is a zip file containing the .npy casual order adjency matrix.
Contact: ml4ed @ UMass Amherst
- Nischal Ashok Kumar ([email protected])
- Wanyong Feng ([email protected])
- Jaewook Lee ([email protected])
- Hunter McNichols ([email protected])
- Aritra Ghosh ([email protected])