/riiid-aied-challenge

Explores classification methods to predict whether a student will answer a test question correctly

Primary LanguageJupyter Notebook

riiid-aied-challenge

Made withJupyter python (scoped)

Description

riiid-aied-challenge explores classification methods to predict whether a student will answer a test question correctly. riiid-aied-challenge attemps to solve the Riiid AIEd Challenge that was presented as the Kaggle competition: Riiid! Answer Correctness Prediction.

Usage

Environment

Navigate to the riiid-aied-challenge directory and setup a new conda environment using the following commands.

conda create -n riiid python=3.8.5 -y
conda activate riiid
conda install ipykernel -y
ipython kernel install --user --name=riiid

Dependencies

Install the dependencies using the following command.

pip install -r requirements.txt

Execution

To train and test the classifiers, run the cells of the Jupyter notebooks, using jupyter lab, ensuring the riiid kernel is selected.

Training data should be placed at data/train.csv. The training data CSV used by riiid-aied-challenge can be found here.

Results

The following are loss vs. epoch plots for the classifier's training stages:

Binary Classifier (3 Layers with Batch Normalization)

Binary Classifier (4 Layers with Batch Normalization)

Binary Classifier (3 Layers without Batch Normalization)

LSTM Binary Classifier (3 LSTM layers, 2 FC layers)

Authors

  • Rishi Masand
  • Arjun Arun

Resources

PyTorch [Tabular] — Binary Classification by Akshaj Verma

CNN-LSTM PyTorch RIIID by Shivanand M N