This is the top1 solution in task 2 of the NeurIPS 2020 Education Challenge
requirements:
Linux,
Python 3.7,
tensorflow 2.3.0,
Scikit-learn 0.23.2
Firstly, download the data from the competiton website, then run following code in order:
- data preprocess
question_meata_change.py: reindex the knowledge tags (knowledge concepts)
answer_meta_change.py: change time to the form of continuous integers timestamp
task_12_change.py: add two new columns, timestamp and quizid, to the traning data
- generate data
task2_data_quiz.py: generate the data for traning and testing. The answered records are sorted by quizid and timestamp, sequence length is 10
data_split_task2.py: split 10-fold traing data
- ocd_model
model training: python train.py kfold batch_size
model testing: python test.py kfold model_name
The model architecture is below:
