Posit (RStudio)¶

# Install from CRAN
install.packages("ellmer")

# Or install development version from GitHub
# install.packages("pak")
pak::pak("tidyverse/ellmer")

library(ellmer)

# Create a chat with OpenAI
chat <- chat_openai(
  model = "gpt-4o-mini",
  system_prompt = "You are a helpful data science assistant."
)

# Have a conversation
chat$chat("What is the tidyverse?")

# Continue the conversation (context is retained)
chat$chat("Which packages are included?")

# View token usage
chat$token_usage()

library(ellmer)

# Create a chat with Claude
chat <- chat_anthropic(
  model = "claude-sonnet-4-5-20250514",
  system_prompt = "You are an expert R programmer."
)

# Ask about R code
chat$chat("How do I read a CSV file with readr?")

library(ellmer)

chat <- chat_openai(model = "gpt-4o-mini")

# Stream the response to the console
chat$chat("Explain the central limit theorem", echo = "all")

library(ellmer)

# Define the structure you want to extract
paper_info <- type_object(
  title = type_string("The paper title"),
  authors = type_array(type_string("Author names")),
  year = type_integer("Publication year"),
  journal = type_string("Journal name")
)

chat <- chat_openai(model = "gpt-4o-mini")

# Extract structured information
result <- chat$extract_data(
  "Smith, J., & Jones, M. (2024). Machine learning in ecology.
   Nature Ecology & Evolution, 8(3), 234-245.",
  type = paper_info
)

print(result)
# $title
# [1] "Machine learning in ecology"
#
# $authors
# [1] "Smith, J." "Jones, M."
#
# $year
# [1] 2024
#
# $journal
# [1] "Nature Ecology & Evolution"

library(ellmer)

# Define a tool
get_weather <- function(city) {
  # In practice, this would call a weather API
  paste("The weather in", city, "is sunny and 72F")
}

chat <- chat_openai(
  model = "gpt-4o-mini",
  system_prompt = "You help users with weather information."
)

# Register the tool
chat$register_tool(
  name = "get_weather",
  description = "Get the current weather for a city",
  arguments = list(
    city = type_string("The city name")
  ),
  fn = get_weather
)

# The model will call the function when appropriate
chat$chat("What's the weather like in Phoenix?")

library(ellmer)

# Test OpenAI
tryCatch({
  chat <- chat_openai(model = "gpt-4o-mini")
  chat$chat("Say hello!")
  message("OpenAI API key is working!")
}, error = function(e) {
  message("OpenAI API key issue: ", e$message)
})

# Test Anthropic
tryCatch({
  chat <- chat_anthropic(model = "claude-sonnet-4-5-20250514")
  chat$chat("Say hello!")
  message("Anthropic API key is working!")
}, error = function(e) {
  message("Anthropic API key issue: ", e$message)
})

library(ellmer)

# List available Ollama models
models_ollama()

# Create a chat with a local model
chat <- chat_ollama(
  model = "llama3.2",
  system_prompt = "You are a helpful research assistant."
)

# Chat with the local model
chat$chat("Explain the difference between correlation and causation")

# Connect to Ollama on a different host
chat <- chat_ollama(
  model = "llama3.2",
  base_url = "http://192.168.1.100:11434"  # Remote Ollama server
)

chat <- chat_ollama(
  model = "llama3.2",
  system_prompt = "You are a statistics tutor.",
  params = params(
    temperature = 0.7,  # Creativity (0-2)
    num_ctx = 4096      # Context window size
  )
)

library(ellmer)
library(tidyverse)

# 1. Create an Ollama chat for data analysis assistance
analyst <- chat_ollama(
  model = "llama3.2",
  system_prompt = "You are a data analysis expert. Provide R code examples
                   using tidyverse packages. Be concise and practical."
)

# 2. Get help with data manipulation
analyst$chat("How do I group data by multiple columns and calculate
              summary statistics in dplyr?")

# 3. The model provides code - let's use it
mtcars %>%
  group_by(cyl, gear) %>%
  summarise(
    mean_mpg = mean(mpg),
    sd_mpg = sd(mpg),
    n = n(),
    .groups = "drop"
  )

# 4. Ask follow-up questions
analyst$chat("How can I visualize this with ggplot2?")

library(ellmer)
library(tidyverse)

# Create a specialized assistant
lit_review <- chat_ollama(
  model = "qwen2.5:7b",
  system_prompt = "You are an academic research assistant specializing in
                   literature reviews. Help researchers:
                   1. Identify key themes in abstracts
                   2. Suggest search terms
                   3. Evaluate methodology descriptions
                   Be scholarly but accessible."
)

# Analyze an abstract
abstract <- "This study examines the impact of social media use on
adolescent mental health using a longitudinal design with 5,000 participants
over 3 years. We found significant associations between daily social media
time and symptoms of anxiety and depression, with effect sizes varying by
platform type and usage patterns."

lit_review$chat(paste("Analyze this abstract and identify:",
                      "1. Research question",
                      "2. Methodology strengths",
                      "3. Potential limitations",
                      "4. Related search terms for similar studies",
                      "\n\nAbstract:", abstract))

library(ellmer)

# Create a statistics tutor
stats_help <- chat_ollama(
  model = "llama3.2",
  system_prompt = "You are a patient statistics tutor for graduate students.
                   When explaining:
                   1. Start with intuition before formulas
                   2. Use concrete examples
                   3. Show R code for implementation
                   4. Explain when methods are appropriate"
)

# Get help choosing a statistical test
stats_help$chat("I have a dataset with one continuous outcome variable and
                 two categorical predictors (treatment group with 3 levels,
                 and gender with 2 levels). I want to understand if there's
                 an interaction effect. What test should I use and why?")

library(ellmer)

# Create a documentation assistant
doc_helper <- chat_openai(
  model = "gpt-4o-mini",
  system_prompt = "You are an R documentation expert. When given R code,
                   generate roxygen2-style documentation including:
                   @title, @description, @param, @return, @examples"
)

# Generate documentation for a function
my_function <- "
calculate_effect_size <- function(group1, group2, pooled_sd = TRUE) {
  n1 <- length(group1)
  n2 <- length(group2)
  m1 <- mean(group1)
  m2 <- mean(group2)

  if (pooled_sd) {
    s <- sqrt(((n1-1)*var(group1) + (n2-1)*var(group2)) / (n1+n2-2))
  } else {
    s <- sd(c(group1, group2))
  }

  d <- (m1 - m2) / s
  return(list(cohens_d = d, n1 = n1, n2 = n2))
}
"

doc_helper$chat(paste("Generate roxygen2 documentation for this R function:\n",
                      my_function))

# install.packages("tidyllm")
library(tidyllm)

llm_message("Explain linear regression") |>
  openai_chat(model = "gpt-4o-mini") |>
  get_reply()

# install.packages("gptstudio")
library(gptstudio)

# Provides RStudio addins for:
# - Code explanation
# - Documentation writing
# - Code commenting
# - Spelling/grammar checking

# install.packages("chattr")
library(chattr)

# Opens an interactive chat pane in RStudio
chattr_app()

# install.packages("ollamar")
library(ollamar)

# Low-level Ollama API access
list_models()
generate("llama3.2", "Hello!")