Mistakes that data science students make

🌈 Abstract

The article discusses the common mistakes and challenges faced by students in introductory data science programming courses, based on an analysis of student code submissions in a data science course at the University of Michigan.

🙋 Q&A

[01] What kind of problems do students face in introductory data science courses?

• Logical errors due to misunderstanding the data or problem statement, such as using the wrong columns/values or incorrectly handling missing values
• Semantic mistakes in using the incorrect function or operator, which may or may not throw a runtime error
• Inefficient code that does not use best practices, such as using for loops instead of vectorized operations
• Misconceptions about Python or Jupyter notebooks, such as incorrectly specifying paths, using incorrect syntax, or having scoping issues

[02] What are the key competencies for introductory data science courses?

• Programming in computational notebooks
• Data literacy and understanding the dataset domain
• Programming with data science libraries
• Programming strategies
• Code optimality

[03] How did the researchers try to address these challenges?

• They deployed an LLM-powered AI tutor in a large data science course, which aimed to help students not just with code correctness, but also domain knowledge, data literacy, and data science best practices.
• The system used the student's code, program output, AST information, characteristics of the dataframe, and code optimality metrics, as well as previously graded assignments from past semesters of the course to provide better feedback.