jbarrow / reteach

Duolingo SLAM experiments, written in AllenNLP

reteach

reteach is a new approach to deep knowledge tracing by considering:

1. the timeline of the user studying each individual knowledge concept; and
2. the interaction between different concepts.

By taking this approach, we hope to build o

Getting Started

Installation

To install reteach, you must first install the requirements:

pip install -r requirements.txt

Then, just clone the repo and you should be good to go.

Data

The original data can be found on the Duolingo 2018 SLAM Shared Task page. The validation and test datasets were blind and thus came with labels in a separate file. To join the labels with the data, just run our join_labels.py script:

python -m scripts.join_labels [PATH_TO_DATA_FILE] [PATH_TO_KEY_FILE] > [PATH_TO_OUTPUT_FILE]

Training

To train a model, run:

cuda=0 allennlp train \
  -f \
  --include-package reteach \
  -s [TRAINED_MODEL_DIRECTORY] \
  conf/slam.jsonnet

Inference

You can generate a prediction file (with JSON outputs) by running the following command:

allennlp predict \
  --output-file [OUTPUT_FILE] \
  --silent \
  --include-package reteach \
  --predictor slam-predictor \
  --use-dataset-reader \
  [TRAINED_MODEL_DIRECTORY] \
  [DATASET]