Hello everyone! I hope you had a great holiday and here's to a fantastic 2024! In my first blog of the year, I want to tell you about my Outreachy project, where I'm working on a machine learning model to find accessibility bugs on Mozilla's Bugzilla system. It all started on an interesting GitHub repository called Bugbug.
Bugbug is a platform for machine learning projects in software engineering. It's got models, data, and scripts for handling various tasks on bugs from Bugzilla, like bug triage, regression, duplicate detection, spam bug detection and component assignment. When my model goes live, it'll be one of Bugbug's many models, specifically geared towards figuring out if a reported bug is an accessibility issue.
The classifier for the model is an XGBoost Classifier from XGBoost, an open-source library that's great for machine learning. The model goes through training and testing on past bugs tagged with the 'access' keyword and Accessibility Severity field on Bugzilla. It takes a bug report as input and gives out a probability score, indicating how likely it is that the bug is an accessibility problem. The model works alongside Bugbot, which runs a rule to add the keyword "access" to a bug if the model predicts it's an accessibility bug, with a certain level of confidence.
This project is exciting because it's pushing me to learn new things like gradient boosting, feature engineering, and model evaluation. I get to play around with tools like Python, XGBoost, Scikit Learn, and dive into the world of Machine Learning. More importantly, I've learned how crucial it is to choose the right features in machine learning. Of course, there have been challenges, like figuring out how to collect, process, and roll back data from Bugzilla, tuning XGBoost parameters, and making sense of the model results.
As I keep going on this journey, I'm looking forward to polishing my model, taking on new challenges, and hopefully, making a positive impact on how accessibility is handled in Bugzilla. Right now, I'm waiting for it to be integrated into Bugbot, so stay tuned. If you ever see an accessibility bug on Bugzilla automatically tagged, you might just be witnessing one of my proudest contributions to the open-source community. Cheers!
Comments
Post a Comment