Fine-Grained Knowledge Tracing model (FGKT).
This project is the Pytorch implementation for FGKT.

If you have more questions about our experiments, you can contact us. email: [email protected]

In 'data' folder, we have provided the processed datasets. If you would like to access the raw datasets, the raw datasets are placed in the following links:

• Statics : address
• Synthetic-5 : address
• ASSISTments2009 : address
• ASSISTments2015 : address
• ASSISTments Competition : address

Service:

• Linux operation system

Environment:

• python 3+
• sklearn 0.21.3
• tqdm 4.54.1
• torch 1.7.0
• numpy 1.19.2

Here is a example for using FGKT model (on ASSISTments Competition):

  python main.py --dataset assist2017  

If you do not want to use the default parameters for your experiments, you can change the model parameters in the following way:

  python main.py --dataset assist2017 --gpu 0 --patience 5 --lr 0.001 --num_heads 1  --mode 3 --exercise_embed_dim 128 --batch_size 32

Explanation of parameters:

• gpu: Specify the GPU to be used, e.g '0,1,2,3'. If CPU is used then fill in -1.
• patience: Maximum number of times if validation loss does not decrease.
• lr: Learning rate
• num_heads: Number of head attentions.
• mode: Selection of integration function.
• exercise_embed_dim: Number of exercise embedding dimensions.
• batch_size: Number of batch size.