Week 00: AI-Assisted Coding Prep

Boosting coding for data analysis with AI tools

Published

June 1, 2025

Boosting coding for data analysis with AI tools

Learning Objectives

By the end of this session, students will:

Understand how AI transforms coding workflows for data analysis tasks
Learn effective prompting strategies for data analysis code generation
Experience key AI coding tools: ChatGPT Canvas, Claude Projects, GitHub Copilot
Practice AI-assisted coding for common data analysis tasks
Set up integrated development environments with AI assistance
Focus on R and Python.

Preparation / Before Class

🔧 Tool Setup

Required Access:

ChatGPT account (free tier sufficient for testing)
Claude account (free tier sufficient)

Recommended Setup:

GitHub account for Copilot (free for students)
VSCode, RStudio, or Jupyter Notebook installed
Your preferred data analysis language ready (R/Python)
You can you a great deal of this for other languages like SQL, Stata or Julia.

Optional but Valuable:

Cursor AI editor (free trial available)

Class Material

🤖 Why AI + Data Analysis Coding Works (20 min)

Covers key set of tasks:

Repetitive patterns: Data analysis has common workflows (load → clean/wrangle → analyze → visualize)
Well-documented libraries: pandas, scikit-learn, as well as dplyr, ggplot2, are extensively covered in AI training data (had been in use for a while). With updates, it knows more recent libraries from tensorflow to fixest (R)
Clear intentions: “Create a scatterplot with regression line” is specific enough for good code generation. Well established “good practice”
Iterative nature: Data analysis involves lots of tweaking and refinement – chat aspect helpful.

What AI Excels At:

Well defined code chunks and setup (libraries, dependencies)
Syntax, how exactly do certain coding tasks (such as regex, loops)
Standard statistical procedures
Data manipulation and cleaning
Basic visualization code

Human Oversight and Decisions are Needed:

Most code
Research design decisions
Statistical / econometric interpretation
Domain-specific logic
Finalize question (vs bland suggestions)
Quality control and validation

📝 Effective Prompting for Data Analysis (25 min)

Model Recommendations:

ChatGPT 4o / Claude Sonnet 4. works about equally fine.

Prompting Best Practices

Some ideas to help get the code do what you want, or at least close enough.

Be specific

Be specific about libraries: “Using pandas and seaborn…” vs. “Using R and ggplot2…”
Include data structure: “DataFrame with columns: date, price, volume”
Specify output format: “Save as PNG for publication” or “Return as tidy data table” or “give back markedown (latex) text with equations”

Define language, preferences

I use R and tidyverse, so unless specified, use that
I use Python, and when possible prefer polars to pandas
Ask for comments: “Include detailed comments explaining each step”

For frequent tasks, AI will know which library to use in a language

Filter on size<5 – will do it in Python Pandas as default.
Scatterplot of sales and employment – Will do it seaborn / matplotlib or ggplot in R.

Advice include data structure

If possible upload the data
if not, upload a small bit of data, like 1/1000 random sample
Start with creating a data dictionary (see also class 3)

Example prompts

You can experiment with a vague prompt or being specific.

Example Prompt 1: broad

"Here is some sales data, summarize regional variation."

Example Prompt 2: with details

"Create Python (R) code using pandas (dplyr) to:

1) Load CSV with columns id, date, sales, region
2) Filter for 2023 data  
3) Group by region and calculate mean sales
4) Create a bar chart with plotnine / ggplot2"

What are the pros and cons of each?

Some tasks where AI works well

1. Data Cleaning Pipeline 2. Exploratory Data Analysis 3. Statistical Modeling 4. Machine Learning Pipeline

📝 Creating Graphs exercise (40 min)

A full practice exercise

Created one example on gender wage gap.

It follows the case study on the in Chapter 09

We start with actual earnings data from the US, called CPS MORG.

We end up with:

Other ideas

Instructors may have their datasets and exercises. We have plenty here: Case studies, each comes with data

Setting up tool

🛠️ AI Coding Tools Showcase (30 min)

ChatGPT Tools:

Canvas: Collaborative coding environment for iterative development
Advanced Data Analysis: Upload datasets, generate analysis with code execution
GPT-4.1: Optimized specifically for coding tasks

Claude Tools:

Projects: Upload full datasets and documentation for context-aware coding
Artifacts: Code generation with real-time preview and editing
Claude 4 Sonnet: Strong reasoning for complex analytical workflows

Specialized Coding Tools:

GitHub Copilot: Inline code completion integrated into your existing editor
Cursor AI: AI-first code editor with context-aware suggestions
Replit: Browser-based coding with AI assistance

Hands-on Demo:

Upload sample dataset to ChatGPT Advanced Data Analysis
Create Claude Project with course data
Compare code generation approaches

⚡ GitHub Copilot Integration (20 min)

Setup in Different Environments:

VSCode:

Install GitHub Copilot extension
Authenticate with GitHub account
Use Ctrl+Space for suggestions, Tab to accept

Jupyter Notebook:

Install via VS Code Jupyter extension or JupyterLab extension
Inline suggestions while typing
Copilot Chat for longer explanations

RStudio:

Enable GitHub Copilot in Global Options > Code > Completion
Works with R scripts and R Markdown
Suggests tidyverse and base R patterns

Workflow Best Practices:

Write descriptive comments before code blocks
Use meaningful variable names to guide suggestions
Accept suggestions, then modify as needed
Use Copilot Chat for explanations and debugging

Discussion Questions

Reflection:

Which AI tool felt most natural for your coding style?
Where did AI suggestions surprise you (positively or negatively)?
How might this change your typical data analysis workflow?
What validation steps would you add when using AI-generated code?

Background, Tools and Resources

Getting Started:

GitHub Student Pack – Free Copilot access
Cursor AI – AI-first code editor
OpenAI Codex Cookbook – Advanced prompting examples

Key Insight: AI coding assistance is most powerful when you understand the underlying concepts. Use AI to accelerate implementation, not replace understanding.

Next Week: Week 1 - LLM Review where we’ll explore broader AI concepts for data analysis.

Some personal comments on AI and this class

While this class is called week00, it was created last. You may have guessed righ.
AI (Claude 4.0) created a great deal of this class following a detailed prompt. I also asked whether to keep it as week00 or change all numbering. Great answers.

--- title: "Week 00: AI-Assisted Coding Prep" subtitle: "Boosting coding for data analysis with AI tools" date: "2025-06-01" --- :::::: hero-section ::::: container ::: hero-title Week 00: AI-Assisted Coding Prep ::: ::: hero-subtitle Boosting coding for data analysis with AI tools ::: ::::: :::::: ------------------------------------------------------------------------ ![](../images/week0_pic.png) ## Learning Objectives By the end of this session, students will: - Understand how AI transforms coding workflows for data analysis tasks - Learn effective prompting strategies for data analysis code generation - Experience key AI coding tools: ChatGPT Canvas, Claude Projects, GitHub Copilot - Practice AI-assisted coding for common data analysis tasks - Set up integrated development environments with AI assistance - Focus on R and Python. ## Preparation / Before Class ::::: {.week-card .card} ::: card-header 🔧 **Tool Setup** ::: ::: card-body **Required Access:** - ChatGPT account (free tier sufficient for testing) - Claude account (free tier sufficient) **Recommended Setup:** - GitHub account for Copilot (free for students) - VSCode, RStudio, or Jupyter Notebook installed - Your preferred data analysis language ready (R/Python) - You can you a great deal of this for other languages like SQL, Stata or Julia. **Optional but Valuable:** - Cursor AI editor (free trial available) ::: ::::: ## Class Material ::::: {.week-card .card} ::: card-header 🤖 **Why AI + Data Analysis Coding Works (20 min)** ::: ::: card-body **Covers key set of tasks:** - **Repetitive patterns:** Data analysis has common workflows (load → clean/wrangle → analyze → visualize) - **Well-documented libraries:** pandas, scikit-learn, as well as dplyr, ggplot2, are extensively covered in AI training data (had been in use for a while). With updates, it knows more recent libraries from tensorflow to fixest (R) - **Clear intentions:** "Create a scatterplot with regression line" is specific enough for good code generation. Well established "good practice" - **Iterative nature:** Data analysis involves lots of tweaking and refinement -- chat aspect helpful. **What AI Excels At:** - Well defined code chunks and setup (libraries, dependencies) - Syntax, how exactly do certain coding tasks (such as regex, loops) - Standard statistical procedures - Data manipulation and cleaning - Basic visualization code **Human Oversight and Decisions are Needed:** - Most code - Research design decisions - Statistical / econometric interpretation - Domain-specific logic - Finalize question (vs bland suggestions) - Quality control and validation ::: ::::: ::::: {.week-card .card} ::: card-header 📝 **Effective Prompting for Data Analysis (25 min)** ::: ::: card-body **Model Recommendations:** * ChatGPT 4o / Claude Sonnet 4. works about equally fine. ### Prompting Best Practices Some ideas to help get the code do what you want, or at least close enough. **Be specific** - **Be specific about libraries:** "Using pandas and seaborn..." vs. "Using R and ggplot2..." - **Include data structure:** "DataFrame with columns: date, price, volume" - **Specify output format:** "Save as PNG for publication" or "Return as tidy data table" or "give back markedown (latex) text with equations" **Define language, preferences** - I use R and tidyverse, so unless specified, use that - I use Python, and when possible prefer polars to pandas - Ask for comments: "Include detailed comments explaining each step" **For frequent tasks, AI will know which library to use in a language** - **Filter on size<5** -- will do it in Python Pandas as default. - **Scatterplot of sales and employment** -- Will do it seaborn / matplotlib or ggplot in R. **Advice include data structure** - If possible upload the data - if not, upload a small bit of data, like 1/1000 random sample - Start with creating a data dictionary (see also [class 3](/week03)) ### Example prompts You can experiment with a vague prompt or being specific. **Example Prompt 1: broad** ``` "Here is some sales data, summarize regional variation." ``` vs **Example Prompt 2: with details** ``` "Create Python (R) code using pandas (dplyr) to: 1) Load CSV with columns id, date, sales, region 2) Filter for 2023 data 3) Group by region and calculate mean sales 4) Create a bar chart with plotnine / ggplot2" ``` What are the pros and cons of each? ### Some tasks where AI works well *1. Data Cleaning Pipeline *2. Exploratory Data Analysis *3. Statistical Modeling *4. Machine Learning Pipeline ::: ::::: ::::: {.week-card .card} ::: card-header 📝 **Creating Graphs exercise (40 min)** ::: ::: card-body **A full practice exercise** Created one example on [gender wage gap](/week00/assets/creating-graphs.html). It follows the case study on the [in Chapter 09](https://gabors-data-analysis.com/casestudies/#ch09a-estimating-gender-and-age-differences-in-earnings) We start with actual earnings data from the US, called CPS MORG. We end up with: ![Nice heatmap](../images/earnings-heatmap.png) **Other ideas** * Instructors may have their datasets and exercises. We have plenty here: [Case studies](https://gabors-data-analysis.com/casestudies/), each comes with [data](https://osf.io/7epdj/) ::: ::::: ## Setting up tool ::::: {.week-card .card} ::: card-header 🛠️ **AI Coding Tools Showcase (30 min)** ::: ::: card-body **ChatGPT Tools:** - **Canvas:** Collaborative coding environment for iterative development - **Advanced Data Analysis:** Upload datasets, generate analysis with code execution - **GPT-4.1:** Optimized specifically for coding tasks **Claude Tools:** - **Projects:** Upload full datasets and documentation for context-aware coding - **Artifacts:** Code generation with real-time preview and editing - **Claude 4 Sonnet:** Strong reasoning for complex analytical workflows **Specialized Coding Tools:** - **GitHub Copilot:** Inline code completion integrated into your existing editor - **Cursor AI:** AI-first code editor with context-aware suggestions - **Replit:** Browser-based coding with AI assistance **Hands-on Demo:** - Upload sample dataset to ChatGPT Advanced Data Analysis - Create Claude Project with course data - Compare code generation approaches ::: ::::: ::::: {.week-card .card} ::: card-header ⚡ **GitHub Copilot Integration (20 min)** ::: ::: card-body **Setup in Different Environments:** **VSCode:** - Install GitHub Copilot extension - Authenticate with GitHub account - Use Ctrl+Space for suggestions, Tab to accept **Jupyter Notebook:** - Install via VS Code Jupyter extension or JupyterLab extension - Inline suggestions while typing - Copilot Chat for longer explanations **RStudio:** - Enable GitHub Copilot in Global Options \> Code \> Completion - Works with R scripts and R Markdown - Suggests tidyverse and base R patterns **Workflow Best Practices:** - Write descriptive comments before code blocks - Use meaningful variable names to guide suggestions - Accept suggestions, then modify as needed - Use Copilot Chat for explanations and debugging ::: ::::: ## Discussion Questions **Reflection:** - Which AI tool felt most natural for your coding style? - Where did AI suggestions surprise you (positively or negatively)? - How might this change your typical data analysis workflow? - What validation steps would you add when using AI-generated code? ## Background, Tools and Resources **Getting Started:** - [GitHub Student Pack](https://education.github.com/pack) -- Free Copilot access - [Cursor AI](https://www.cursor.com/) -- AI-first code editor - [OpenAI Codex Cookbook](https://github.com/openai/openai-cookbook) -- Advanced prompting examples **Key Insight:** AI coding assistance is most powerful when you understand the underlying concepts. Use AI to accelerate implementation, not replace understanding. **Next Week:** [Week 1 - LLM Review](../week01/) where we'll explore broader AI concepts for data analysis. ## Some personal comments on AI and this class - While this class is called week00, it was created last. You may have guessed righ. - AI (Claude 4.0) created a great deal of this class following a [detailed prompt](assets/week00-ask-ai). I also asked whether to keep it as week00 or change all numbering. Great answers.