Talks # 15: Shubhadeep Roychowdhury; Applying Machine Learning on Source Code | Summary and Q&A

TL;DR

Learn how machine learning can improve code documentation by automatically generating docstrings and providing type checking and bug detection.

Key Insights

• 👨‍💻 Machine learning can enhance code documentation by generating docstrings, providing type checking, and detecting bugs.
• ℹ️ Open-source libraries like Treehugger by Codist offer a unified API to analyze different programming languages and extract useful information from code.
• 👨‍💻 Companies like Kite, TabNine, and Microsoft are actively using machine learning to improve code completion and documentation.
• 👨‍💻 Limitations of machine learning on source code include the lack of interpretability, the challenge of an open vocabulary in code, and the need for a fusion of deep learning and symbolic AI approaches.

Transcript

my i'm perfectly audible right i remember my voices hello everyone and welcome to the new episode of talks today we have shuba deep roy chaudhary he is cto owner and co-founder of codist it's a paris bait startup and they have created dockly which is a automated code summarization tool and today he is going to talk about how to document your code b... Read More

Questions & Answers

Q: How can machine learning be applied to code documentation?

Machine learning models can be used to generate docstrings for code functions, provide type checking, detect bugs, and even generate unit tests. These models learn from large code corpora to predict useful information about code.

Q: What are some real-world applications of machine learning on source code?

Companies like Kite, TabNine, and Microsoft are using machine learning to enhance code completion and documentation. For example, Kite offers an auto-completion engine integrated in VS Code, while TabNine provides code auto-completion using machine learning. Microsoft's Visual Studio IntelliCode also uses machine learning for AI-assisted development.

Q: What are the limitations of machine learning on source code?

One limitation is the lack of common sense and interpretability in deep learning models. They may struggle to understand concepts specific to code, such as arithmetic operations. Another limitation is the challenge of dealing with open vocabulary in code, where different developers may use different naming conventions and code styles.

Q: How does Dokley, by Codist, improve code documentation?

Dokley is a tool developed by Codist that automatically generates docstrings for Python code. It uses a machine learning model to predict the purpose and functionality of code functions, helping developers write better documentation.

Summary & Key Takeaways

• Machine learning can be applied to source code to automate code summarization, generate docstrings, and provide type checking and bug detection.

• Companies like Kite, TabNine, and Microsoft are already using machine learning to enhance code completion and documentation.

• Treehugger, an open-source library by Codist, provides a unified API to analyze different programming languages and extract useful information from code.