Going Deeper with Deep Knowledge Tracing - EDM-2016

Source code and data sets for Going Deeper with Deep Knowledge Tracing

Dependencies:

• Tensorflow 0.10 (only tested on this version)
• Numpy
• scikit-learn

Get Started

git clone https://github.com/siyuanzhao/2016-EDM.git
cd 2016-EDM
python student_model.py

Usage

There are serval flags within student_model.py. Some of them are hyper-parameters for the model. Some of them are path to training and testing data.

Check all available flags with the following command.

python student_model.py -h

Run the model on a different data set

python student_model.py --train_data_path=<path-to-your-data> --test_data_path=<path-to-your-data>

You can also set the number of hidden layers and the number of hidden nodes with flags.

python student_model.py --hidden_layer_num=2 --hidden_size=400

Details

• The model uses Adam Optimizer.
• Add gradient noise. arxiv
• Add gradient norm clipping. arxiv

Datasets

Data sets we used from the paper are in folder data.

	ASSISTments 09-10 (a)	ASSISTments 09-10 (b)	ASSISTments 09-10 (c)
file name	0910_a	0910_b	0910_c
Has duplicated records	No	No	No
Has subproblems	Yes	No	No
Repeat sequnces for mutiple skills	Yes	Yes	No
Combined skills for mutiple skills	No	No	Yes

CAT_train.csv and CAT_test.csv are data files from Cognitive Tutor.

Results

Since I made some changes on the code, I will run the model again and record the results.