Back

PyART API Recommendation

Using ML Models to Make Programming More Accessible...

This project utilized various machine learning models to improve a Python API recommendation model. I used Random Forest, SVM, Naive Bayes, Logistic Regression, and Decision Tree to improve the models. I then tested it on various Python projects, ranging from CleverCSV to marshmallow to establish how accurately it could predict APIs. I also used Understand to collect JSON files of the functions in a project. I tested around 15 projects with all of the models and used Top-K accuracy and MRR to deduce the accuracies of the models. The best-suited model often depended on the project.

The results of this project highlight that while these models are so incredibly valuable in integrating ML into important aspects of day-to-day, models made from scratch have the potential to take it a step further. I hope to explore creating tailored machine learning models in the future, especially since an accurate API Recommendation model could change the programming game, making it far more efficient and accessible. Some platforms such as GitHub Copilot are integrating this type of technology, but it's not always accurate. Still, it's an immense step forward considering that it makes programming more accessible overall.

Source Project
Project Recommendation Results
Top-K Accuracy