Authors: Aral de Moor, Arie van Deursen, Maliheh Izadi

Published on: May 23, 2024

Impact Score: 7.6

Arxiv code: Arxiv:2405.14753

Summary

• What is new: Developed a machine learning model to predict when to provide code completion suggestions based on code context and telemetry data.
• Why this is important: Existing transformer models for code completion are effective but expensive and disruptive due to frequent, non-contextual suggestions.
• What the research proposes: A new model that uses developer interactions and telemetry data to better predict optimal times for code completion suggestions.
• Results: The new model outperforms existing models in accuracy and latency, with successful real-world deployment and positive feedback from developers.

Technical Details

Technological frameworks used: Transformer-based frameworks

Models used: Small-scale transformer models for invocation filtering

Data used: 200k developer interactions dataset and telemetry data

Potential Impact

Software development tools, IDE plugin markets, and companies developing coding assistance tools

Want to implement this idea in a business?

We have generated a startup concept here: CodeFlow.