Classify and visualize {ellmer} git commits

ellmer-commits.R

library(gh)
library(dplyr)
library(hellmer)
library(ggplot2)
library(gganimate)
library(dplyr)
library(tibble)
library(lubridate)
library(scales)
library(cli)
library(tidyr)

# classification

get_commits <- function(owner, repo) {
  page <- 1
  all_commits <- list()
  repeat {
    new_commits <- gh("GET /repos/{owner}/{repo}/commits",
      owner = owner,
      repo = repo,
      per_page = 100,
      page = page
    )
    if (length(new_commits) == 0) break
    all_commits <- c(all_commits, new_commits)
    page <- page + 1
  }
  lapply(all_commits, function(commit) {
    data.frame(
      sha = commit$sha,
      author = ifelse(is.null(commit$commit$author$name), NA, commit$commit$author$name),
      date = commit$commit$author$date,
      message = commit$commit$message,
      stringsAsFactors = FALSE
    )
  }) |>
    bind_rows()
}

commits <- get_commits("tidyverse", "ellmer")

chat <- chat_future(chat_openai())

type_commit <- type_object(
  "Classify GitHub commits",
  action = type_enum("The commit type", c(
    "add-feature", "patch-function",
    "breaking-change", "document-edit-rename",
    "improve-UI-or-UX", "validate-test-check-errors"
  ))
)

batch <- chat$batch(
  prompts = commits$message,
  type_spec = type_commit,
  judgements = 1,
  state_path = "commits.rds"
)

dat <- batch$texts()

# viz

commit_types <- lapply(dat, function(x) x$action) |> unlist()

commit_data <- tibble(
  action = commit_types,
  date = as.POSIXct(commits$date, format = "%Y-%m-%dT%H:%M:%SZ")
)
all_dates <- unique(commit_data$date)
all_actions <- unique(commit_data$action)
cumulative_data <- expand.grid(
  date = all_dates,
  action = all_actions
) |>
  as_tibble() |>
  arrange(date, action)

commit_counts <- commit_data |>
  arrange(date) |>
  group_by(date, action) |>
  summarize(daily_count = n(), .groups = "drop") |>
  right_join(cumulative_data, by = c("date", "action")) |>
  replace_na(list(daily_count = 0)) |>
  arrange(action, date) |>
  group_by(action) |>
  mutate(count = as.integer(cumsum(daily_count))) |>
  ungroup()

max_count <- max(commit_counts$count)

action_colors <- c(
  "add-feature" = "#4CAF50",
  "patch-function" = "#2196F3",
  "breaking-change" = "#F44336",
  "document-edit-rename" = "#9C27B0",
  "improve-UI-or-UX" = "#FF9800",
  "validate-test-check-errors" = "#FFEB3B"
)

p <- ggplot(commit_counts, aes(x = reorder(action, count), y = count, fill = action)) +
  geom_col() +
  geom_text(
    aes(label = sprintf("%d", count)),
    hjust = -0.2,
    size = 5
  ) +
  coord_flip(clip = "off") +
  labs(
    title = "{ellmer}: GitHub Commits",
    subtitle = "Date: {format(frame_time, '%b %d, %Y')}",
    x = NULL,
    y = "Number of Commits"
  ) +
  theme_minimal() +
  scale_fill_manual(values = action_colors) +
  scale_y_continuous(
    limits = c(0, max_count * 1.15),
    expand = expansion(mult = c(0, 0))
  ) +
  theme(
    legend.position = "none",
    plot.title = element_text(size = 24, face = "bold"),
    plot.subtitle = element_text(size = 18),
    axis.title.x = element_text(size = 16),
    axis.text.x = element_text(size = 14),
    axis.text.y = element_text(size = 14, face = "bold"),
    panel.grid.major.y = element_blank()
  ) +
  transition_time(date) +
  ease_aes("linear")

animation <- animate(p, nframes = 150, fps = 10, width = 800, height = 500, res = 120)
anim_save("ellmer_commits.gif", animation)