hadley

Using Google's generative APIs: lessons learned

Basics

• The API key wasn't in the shell environment. It lives in ~/.Renviron, so the scripts fall back to reading GOOGLE_API_KEY from there.
• Find model IDs by listing them, not by guessing marketing names: GET https://generativelanguage.googleapis.com/v1beta/models?pageSize=200 (the displayName field maps the two). The ones used:

hadley

May 2019

Statistical language R falls out of the TIOBE index top 20

After having been in the top 20 for about 3 years, statistical language R dropped out this month.

This is quite surprising because the field of statistical programming is still booming, especially thanks to the popularity of data mining and artificial intelligence.

Snapshot

hadley

Reverse engineering "What Percent Whimsy Are You?"

Quiz: https://www.buzzfeed.com/laughingbunny255/what-percent-whimsy-are-you

The scoring is fully recoverable — the quiz data is embedded as JSON in the page (data.type = "personality", 9 questions, 4 result buckets).

The algorithm

Each question has exactly 4 answers, and each answer is worth points: 1 toward

hadley

LLM for code generation

Currently three main styles:

1. Autocomplete
   • e.g. github copilot, windsurf
   • Inline "ghost text" as you type
   • Sometimes amazingly good; often pretty useless
   • Need to train yourself to ignore spurious suggestions

hadley

Convert <ingredients> to JSON using the following format:

• If an ingredient has both weight and volume, extract only the weight:

  ¾ cup (150g) dark brown sugar
  [{"name": "dark brown sugar", "quantity": 150, "unit": "g"}]
  

• If an ingredient only lists a volume, extract that.

hadley

library(tidyverse)

# https://twitter.com/buddyherms/status/1576966150680121344 --------------

# PROs: at, by, and regexp examples
# CONs: quite simple

vt_census <- tidycensus::get_decennial(
  geography = "block",
  state = "VT",

hadley

f1 <- function(n) {
  x <- numeric()
  for (i in 1:n) {
    x <- c(x, i)
  }
  x
}

f1.5 <- function(n) {
  x <- numeric()

hadley

data(diamonds, package = "ggplot2")

# Most straightforward
diamonds$ppc <- diamonds$price / diamonds$carat

# Avoid repeating diamonds 
diamonds$ppc <- with(diamonds, price / carat)

# The inspiration for dplyr's mutate
diamonds <- transform(diamonds, ppc = price / carat)

hadley

library(rtweet)
library(tidyverse)

auth_setup_default()
json <- search_tweets("#rstats", n = 5000, include_rts = FALSE, parse = FALSE)[1:5]

tweets <- tibble(json = json) %>% 
  unnest_wider(json) %>% 
  select(statuses) %>% 
  unnest_longer(statuses) %>% 

hadley

library(tidyverse)

simple_data <- tibble(
  group = factor(rep(c("A", "B"), each = 15)),
  subject = 1:30,
  score = c(rnorm(15, 40, 20), rnorm(15, 60, 10))
)

simple_data_se &lt;- simple_data %&gt;%