Instructor Notes
Lesson Design Notes: AI-assisted Coding with Codeium / GitHub Copilot
Users are expected to follow The Carpentries Code of Conduct.
All content is publicly available under the Creative Commons Attribution License.
Lesson Title: AI-assisted Coding with Codeium / GitHub Copilot
[TOC]
Target Audience
- What is their background?
- Researchers in the STEM disciplines with any level of seniority (MSc, PhD, postdocs, faculty staff) and various level of coding expertise, frequently use coding environments like VS Code for data analysis, automation, or simulations. There are also some more experienced coders in the group who require precision and control over their code.
- What do they already know how to do?
- Less experienced coders have minimal coding skills/familiar with any programming language, while more experienced coders are skilled in coding, debugging, and version control. Most have likely heard about AI-assisted tools but may not be fully aware of the best practices for optimizing their use.
- What do they want to do with the skills they will learn from your
lesson?
- Apply the practical skills learned from the lesson to optimize their coding experience, improve the quality and clarity of the code. Less experienced coders want to write better code and get guidance on areas they are unfamiliar with, while experienced coders aim to streamline workflows and code efficiently, managing security and ethical risks effectively.
- What problem will your lesson help them solve?
- Setting up AI coding assistant, code generation, autocompletion, improving existing code, fixing bugs. In particular, it will help less experienced coders identify and avoid errors and write better code, while enabling experienced coders to code more efficiently by automating repetitive tasks and focusing on more complex challenges.
Minimal prerequisites: - Basic computer skills and fundamental coding knowledge (e.g., knowledge of different data types, loading data, writing simple functions) - Familiarity with a code editor (e.g., VSCode) - Interest in AI-assisted coding tools
Lesson Learning Objectives
Learning Objectives
After following this lesson, learners will be able to:
- Generate code and optimize its performance using an AI coding assistant
- Refactor and improve the structure and quality of existing code using an AI coding assistant
- Automate repetitive coding tasks using AI coding assistant
- Generate and improve documentation using AI coding assistant
- Recognize, evaluate, and mitigate ethical and security considerations when using AI coding assistant
Data Set/Narrative
- Which datasets and narratives did you consider?
- We considered dataset which could be interesting for researchers across different fields: climate change, genomics, astronomy.
- How and why did you choose between them?
- We choose climate change, in particular atmosferic CO2 trends, for broad relevance, data complecity and accessibility (see CO2 PPM - Trends in Atmospheric Carbon Dioxide as a tutorial example).
- What implications do you think your choice of dataset and/or
narrative will have for the design and further implementation of your
lesson?
- We will focus on AI-assisted time series analysis, data visualization, and modeling.
Episodes
Episode Learning Objectives
1: Introduction to AI coding assistants - Olga
After following this episode, learners will be able to: * Outline the main AI coding assistants * Set up Codeium as coding assistant for the lesson
2: Code generation and optimization - Olga
After following this episode, learners will be able to: * Generate snippets of code using Codeium as AI-assistant * Optimize performance of the generated code using Codeium as AI-assistant * Refactor and improve the structure and quality of existing code using Codeium as AI-assistant * Identify and fix bugs using Codeium as AI-assistant
3: Enhancing coding efficiency - Giulia
After following this episode, learners will be able to: * Use an autocomplete function of Codeium as AI-assistant to improve coding speed * Automate repetitive coding tasks using Codeium as AI-assistant * Generate and improve documentation using Codeium as AI-assistant * Compose clear and effective prompts that capture and facilitate the expected outcomes using Codeium as AI-assistant
Designing Exercises
Exercise 1: Code and Functions Generation
Objective: Generate snippets of code using Codeium as AI-assistant
Instructions: Using Codeium as AI-assistant: - Read a
datafile into a Pandas DataFrame object, produce datafile
descriptives and plot the data distributions. - Write a function that
takes as input a DataFrame and that calculates basic
descriptive statistics like: number of rows (nrow), number
of columns (ncol), data types of each column, basic summary
statistics (like mean, min, max for numeric columns). - Write a function
that takes as inputs a DataFrame and a column and generate
an histogram to visualize data distribution if the column is numeric
(e.g., int64, float64), a bar plot showing the
category frequency if the column is categorical. - Review the
AI-generated code and compare it to the version you would have written
independently without the AI-assistant.
Solo Activity: Participants work individually on this exercise
Exercise 2: Optimization with Codeium Objective: Use Codeium to improve existing code
Instructions: - Provide a sample code that needs optimization (will be an example from the medical field or, alternatively, I’ll search on stack exchange) - Use Codeium to suggest improvements and restructure the code - Compare the original and restructured versions (clarity, length…)
Solo Activity: Participants work individually on this exercise
Exercise 3: Debugging Assistance Objective: Use Codeium to identify and fix bugs
Instructions: - Introduce a code snippet with errors - Use Codeium to detect and propose fixes - Apply and test the corrections
Group Review: Discuss the debugging process and compare results
Exercise 4: Ethical and Security Challenges Scenario 1: An ethical dilemma involving biased code generation
Scenario 2: A data security issue related to the use of AI tools in a sensitive project
Group Discussion: How participants would address these scenarios
Examples before exercises
Example before exercise 1
Instructions: Using Codeium as AI-assistant: - Create a
function that takes a Pandas DataFrame with a date column
(average), aggregates the monthly data into yearly data,
and then merges this aggregated DataFrame with another
DataFrame that includes a year column. -
Demonstrate how to compare the year column with the added
aggregated values of the average column, and plot them to
ensure they match as expected.
Additional Design Notes
FIXME add notes to this section that do not fit elsewhere in the page. Topics for this section might include
- what has been tried that did not work
- learning objectives that you decided to remove (e.g. to trim down the content) and why
- concept maps for all or part of your lesson (see the section below)
Concept Maps
Concept maps are a useful tool for describing the relationships between concepts. They can be used to visualise one’s mental model of a topic. You can use this section to add concept maps that illustrate the design of your lesson and/or the most important information you are trying to communicate in your lesson/its episodes.
You can embed a photo or other image file, or use the GraphViz syntax demonstrated below.
digraph hierarchy {
nodesep=1.0 // increases the separation between nodes
node [color=darkgray,fontname=Courier,shape=box] //All nodes will this shape and colour
edge [color=navy, style=dashed] //All the lines look like this
"concept map"->{nodes edges} [label="contains"]
edges->nodes [label="connect"]
nodes->concepts [label="describe"]
edges->relationships [label="describe"]
relationships->concepts [label="between"]
{rank=same;nodes edges} // Put them on the same level
"concept map"->"mental model" [label="depicts"]
}Episode Concept Maps
…and concept maps for individual episodes here.
General questions or feedback? Contact team@carpentries.org.