KeyCoin@USTC solution for Global Knowledge Tracing Challenge @AAAI2023

Dependencies:

• python >= 3.8 (or >= 3.8.0 )
• tensorflow-gpu == 2.5.0 (or >= 2.5.0 )
• numpy
• tqdm
• utils
• pandas
• sklearn
• any other necessary package

Hardware:

A GPU with more than 10G memory

Note: the whole training and testing process takes about one and a half hours with the RTX 3090 GPU

pipeline

First, put the datasets (i.e., 'keyid2idx.json', 'pykt_test.csv', 'questions.json', 'train_valid_sequences.csv') in the data file.

Then, make the results folder for saving predictions:

mkdir results

Then, training the model:

python train_all.py

Then, making predictions:

python test_all.py

python meaning.py

The output prediction.csv is our result