[Kaggle] Google AI4Code — Notebook Cell Ordering
Completed

[Kaggle] Google AI4Code — Notebook Cell Ordering

Back to Projects
Python
PyTorch
CodeBERT
DistilBERT
NLP
scikit-learn
Top ~17%
Result
CodeBERT
Model

Figures

Google AI4Code — Understand Code in Python Notebooks. Organised by Google and X, $150,000 prize pool.Open full size
Google AI4Code — Understand Code in Python Notebooks. Organised by Google and X, $150,000 prize pool.

Overview

Google AI4Code challenged competitors to restore the correct cell execution order of shuffled Python Jupyter notebooks — essentially a sequence ordering problem over code and markdown cells.

Task

Predicting the next cell to be executed based on the comments and code in Python Jupyter notebooks.

Approach

Ensemble of CodeBERT and DistilBERT with different weights (0.748 / 0.252) based on the results of running each model. Built on public baseline notebooks, credited in the submission notebook.

Result

Top ~17% on the private leaderboard. Organised by Google and X, $150,000 prize pool.