Creating Effective Prompts for R Coding

A guide to constructing high-quality prompts for leveraging Large Language Models (LLMs) in R.

Fundamental Principles of Prompt Construction

Set Clear Goals

• Specify the task and desired outcome.
• Eliminate ambiguity in the requirements.

Bad Prompt

TASK: Text cleaning function
INPUT: character vector
REQUIREMENTS:
- Remove special characters
- Convert to lowercase
OUTPUT: cleaned vector

Good Prompt

Create an R function that cleans text data by removing special characters and converting to lowercase. The input should be a character vector.

Please provide the code and a brief explanation of your approach.

Use Positive Language

Frame requests in terms of “do this” rather than “don’t do this.” This encourages clarity and actionable guidance.

---

Natural Language and Rigid Formatting

Modern LLMs excel at understanding natural language. Keep prompts conversational but specific.

Provide Context and Specifications

• Specify the desired output format (e.g., R function, script, or data frame).
• Include relevant libraries or packages.

Good Prompt

Using the tidyverse ecosystem in R, write code to analyze my sales data with these specifications:

sales_data <- data.frame(
  date = c("2024-01-01", "2024-01-02", "2024-01-01"),
  region = c("North", "South", "North"),
  sales = c(1200, NA, 1500),
  units = c(50, 45, 60)
)

Requirements:
1. Calculate mean sales and total units by region and month.
2. Format dates appropriately.
3. Return results in a tidied data frame.

Bad Prompt

Average my data in R.

sales_data <- data.frame(
  date = c("2024-01-01", "2024-01-02", "2024-01-01"),
  region = c("North", "South", "North"),
  sales = c(1200, NA, 1500),
  units = c(50, 45, 60)
)

---

Encourage Reasoning and Explanation

• Request step-by-step explanations in addition to code.
• Ask for comments within the code to clarify its functionality.
• Request a summary of the approach or methodology used.

---

Optimize for R-Specific Tasks

Data Manipulation

Specify the preferred libraries or frameworks for data handling, such as dplyr.

Visualization

Explicitly mention visualization libraries like ggplot2 or plotly.

Good Prompt

Create a ggplot2 visualization in R that shows the relationship between `mpg` and `hp` from the mtcars dataset. The plot should:
- Use a scatter plot with a smooth trend line.
- Color the points by the `cyl` variable.
- Use a minimal theme.

---

Improve Code Quality

• Request performance optimization and adherence to established style guides.
• Suggest refactoring to ensure modular, reusable code.

Good Prompt

Refactor the following R code to improve performance and readability using the tidyverse style guide and vector operations.

my_function <- function(df) {
  result <- data.frame()
  for (i in 1:nrow(df)) {
    if (!is.na(df$value[i]) & df$category[i] == "A") {
      temp <- data.frame(
        id = df$id[i],
        transformed = sqrt(df$value[i]),
        group = df$category[i]
      )
      result <- rbind(result, temp)
    }
  }
  return(result)
}

---

Leverage R’s Strengths

• Vectorization: Request the use of vectorized operations where applicable.
• Functional Programming: Encourage using apply functions or the purrr library for functional programming paradigms.
• Data Types: Specify the use of appropriate data structures like lists or data frames to improve efficiency.

---

Iterative Improvement

1. Start with a basic prompt and iterate based on feedback.
2. Review the generated output and refine your requirements.
3. Use specific follow-ups to clarify or improve aspects of the output.
4. Restart with a refined prompt when the direction changes significantly.

---

Summary

Do:

• Clearly specify the required packages and dependencies.
• Include sample data or a detailed description of the expected data structure.
• Request robust error handling and input validation.
• Break down complex tasks into smaller, manageable prompts.
• Ask for explanations of choices or methods used.

Don’t:

• Provide ambiguous or incomplete requirements.
• Overload prompts with unrelated tasks.
• Assume implicit knowledge of specific packages without mentioning them.
• Neglect to specify data types or structures explicitly.

---