CommonLit Readability Prize

4th place solution code for the CommonLit Readability prize hosted on Kaggle (August 2021) https://www.kaggle.com/c/commonlitreadabilityprize

The writeup can be found here: https://www.kaggle.com/c/commonlitreadabilityprize/discussion/258148

Setup

Edit src/config.py to reflect the input and output locations on your machine

Training

To train a single model using a config listed in hyperparams.yml run:

python train.py --config <config_name>

To run a 5-fold cross validation, using 5 different seeds, use the shell script train.sh. This script will also run infer.py and generate out-of-fold (OOF) predictions for stacking models.

sh train.sh <config_name>

Inference

The final submission code that was used for inference in a Kaggle notebook is in the submissions folder

• submission.py - Public RMSE: 0.451, Private RMSE: 0.447
• submission_netflix.py - Public RMSE: 0.452, Private RMSE: 0.446