tidyverse-patterns

现代 tidyverse 模式

使用 dplyr 1.1+ 和 R 4.3+ 进行现代 tidyverse 开发的最佳实践

核心原则

1. 使用现代 tidyverse 模式 — 优先使用 dplyr 1.1+ 特性、原生管道和当前 API
2. 先分析性能，再进行优化 — 使用 profvis 和 bench 找出真正的瓶颈
3. 首先编写易读的代码 — 仅在必要时并且完成性能分析后进行优化
4. 遵循 tidyverse 风格指南 — 保持命名、空格和结构一致

管道用法（|> 而非 %>%）

• 始终使用原生管道 |>，而不是 magrittr %>%
• R 4.3+ 提供了所需的全部功能

# Good - Modern native pipe
data |>
  filter(year >= 2020) |>
  summarise(mean_value = mean(value))

# Avoid - Legacy magrittr pipe
data %>%
  filter(year >= 2020) %>%
  summarise(mean_value = mean(value))

Join 语法（dplyr 1.1+）

• 使用 join_by()，而不是用字符向量指定 Join 条件
• 支持不等式、滚动和重叠 Join

# Good - Modern join syntax
transactions |>
  inner_join(companies, by = join_by(company == id))

# Good - Inequality joins
transactions |>
  inner_join(companies, join_by(company == id, year >= since))

# Good - Rolling joins (closest match)
transactions |>
  inner_join(companies, join_by(company == id, closest(year >= since)))

# Avoid - Old character vector syntax
transactions |>
  inner_join(companies, by = c("company" = "id"))

Join 质量控制

• 使用 relationship 声明基数，以验证 Join 假设
• 使用 unmatched = "error" 捕获意外的未匹配项
• 使用 na_matches = "never"，避免静默的 NA Join
• 交互操作时使用 tidylog:: 前缀，验证 Join 结果

# Validate 1:1 relationship — errors if violated
inner_join(x, y, by = join_by(id),
  relationship = "one-to-one")

# Validate many-to-one (left has duplicates, right does not)
left_join(transactions, companies, by = join_by(company == id),
  relationship = "many-to-one")

# Ensure all rows from left match something in right
inner_join(x, y, by = join_by(id),
  unmatched = "error")

# Prevent NA values from matching each other silently
left_join(x, y, by = join_by(id),
  na_matches = "never")

# Combine for strict joins
inner_join(x, y, by = join_by(id),
  relationship = "one-to-one",
  unmatched = "error",
  na_matches = "never")

# Interactive verification with tidylog
# tidylog prints a summary of rows matched/dropped
tidylog::inner_join(x, y, by = join_by(id))

数据掩码与整洁选择

• 了解数据掩码与整洁选择之间的区别
• 对函数参数使用 {{}}（拥抱）
• 对字符向量使用 .data[[]]

# Data masking functions: arrange(), filter(), mutate(), summarise()
# Tidy selection functions: select(), relocate(), across()

# Function arguments - embrace with {{}}
my_summary <- function(data, group_var, summary_var) {
  data |>
    group_by({{ group_var }}) |>
    summarise(mean_val = mean({{ summary_var }}))
}

# Character vectors - use .data[[]]
for (var in names(mtcars)) {
  mtcars |> count(.data[[var]]) |> print()
}

# Multiple columns - use across()
data |>
  summarise(across({{ summary_vars }}, ~ mean(.x, na.rm = TRUE)))

现代分组和列操作

• 使用 .by 进行逐操作分组（dplyr 1.1+）
• 在数据掩码函数中使用 pick() 选择列
• 使用 across() 对多个列应用函数
• 使用 reframe() 进行多行汇总

# Good - Per-operation grouping (always returns ungrouped)
data |>
  summarise(mean_value = mean(value), .by = category)

# Good - Multiple grouping variables
data |>
  summarise(total = sum(revenue), .by = c(company, year))

# Good - pick() for column selection
data |>
  summarise(
    n_x_cols = ncol(pick(starts_with("x"))),
    n_y_cols = ncol(pick(starts_with("y")))
  )

# Good - across() for applying functions
data |>
  summarise(across(where(is.numeric), mean, .names = "mean_{.col}"), .by = group)

# Good - reframe() for multi-row results
data |>
  reframe(quantiles = quantile(x, c(0.25, 0.5, 0.75)), .by = group)

# Avoid - Old persistent grouping pattern
data |>
  group_by(category) |>
  summarise(mean_value = mean(value)) |>
  ungroup()

NA 安全的行筛选

• 使用 filter_out()，而不是对条件取反 — 取反（!condition）会静默丢弃 NA
• 使用 when_any() 和 when_all() 进行多列 OR/AND 筛选（dplyr 1.2+）

# Problem: negation silently drops rows where condition is NA
filter(data, !(value < 0))       # drops rows where value is NA — silent!

# Good - filter_out() passes NAs through safely
filter_out(data, value < 0)      # rows where value is NA are kept

# Good - when_any() for OR across columns (dplyr 1.2+)
filter(data, when_any(x, y, z, \(col) col > 0))  # any column > 0

# Good - when_all() for AND across columns
filter(data, when_all(x, y, z, \(col) !is.na(col)))  # no NAs in any

# Avoid - verbose base patterns
filter(data, !(value < 0) | is.na(value))   # workaround, not idiomatic

重编码和条件更新

• 使用 replace_when() 进行就地条件更新 — 避免使用 .default = x 的 case_when()
• 所有情况都应处理时，使用带 .unmatched = "error" 的 case_when()

# Good - replace_when() for in-place updates (type-stable, NAs unaffected)
mutate(data, status = replace_when(status,
  value < 0  ~ "negative",
  value == 0 ~ "zero"
))

# Avoid - case_when() requires restating the variable in .default
mutate(data, status = case_when(
  value < 0  ~ "negative",
  value == 0 ~ "zero",
  .default   = status    # repetitive
))

# Good - case_when() with strict exhaustiveness check
mutate(data, grade = case_when(
  score >= 90 ~ "A",
  score >= 80 ~ "B",
  score >= 70 ~ "C",
  .unmatched  = "error"  # error if any row falls through
))

序列化

• 使用 qs2 进行快速序列化 — 它是 qs 的后继版本，且不向后兼容

# Good - qs2 (use .qs2 extension)
qs2::qs_save(object, "data/results.qs2")
object <- qs2::qs_read("data/results.qs2")

# Avoid - older qs package
qs::qsave(object, "data/results.qs")   # outdated

现代 purrr 模式

• 使用 map() |> list_rbind()，而不是已弃用的 map_dfr()
• 使用 walk() 处理副作用（写入文件、绘图）
• 使用 in_parallel() 扩展到多个核心

# Modern data frame row binding (purrr 1.0+)
models <- data_splits |>
  map(\(split) train_model(split)) |>
  list_rbind()  # Replaces map_dfr()

# Column binding
summaries <- data_list |>
  map(\(df) get_summary_stats(df)) |>
  list_cbind()  # Replaces map_dfc()

# Side effects with walk()
plots <- walk2(data_list, plot_names, \(df, name) {
  p <- ggplot(df, aes(x, y)) + geom_point()
  ggsave(name, p)
})

# Parallel processing (purrr 1.1.0+)
library(mirai)
daemons(4)
results <- large_datasets |>
  map(in_parallel(expensive_computation))
daemons(0)

使用 stringr 操作字符串

• 优先使用 stringr，而不是 R 基础字符串函数
• 统一使用 str_ 前缀，并将字符串作为第一个参数
• 专为管道操作和向量化而设计

# Good - stringr (consistent, pipe-friendly)
text |>
  str_to_lower() |>
  str_trim() |>
  str_replace_all("pattern", "replacement") |>
  str_extract("\\d+")

# Common patterns
str_detect(text, "pattern")     # vs grepl("pattern", text)
str_extract(text, "pattern")    # vs complex regmatches()
str_replace_all(text, "a", "b") # vs gsub("a", "b", text)
str_split(text, ",")            # vs strsplit(text, ",")
str_length(text)                # vs nchar(text)
str_sub(text, 1, 5)             # vs substr(text, 1, 5)

# String combination and formatting
str_c("a", "b", "c")            # vs paste0()
str_glue("Hello {name}!")       # templating
str_pad(text, 10, "left")       # padding
str_wrap(text, width = 80)      # text wrapping

# Case conversion
str_to_lower(text)              # vs tolower()
str_to_upper(text)              # vs toupper()
str_to_title(text)              # vs tools::toTitleCase()

# Pattern helpers for clarity
str_detect(text, fixed("$"))    # literal match
str_detect(text, regex("\\d+")) # explicit regex
str_detect(text, coll("e", locale = "fr")) # collation

# Avoid - inconsistent base R functions
grepl("pattern", text)          # argument order varies
regmatches(text, regexpr(...))  # complex extraction
gsub("a", "b", text)           # different arg order

向量化与性能

# Good - vectorized operations
result <- x + y

# Good - Type-stable purrr functions
map_dbl(data, mean)    # always returns double
map_chr(data, class)   # always returns character

# Avoid - Type-unstable base functions
sapply(data, mean)     # might return list or vector

# Avoid - explicit loops for simple operations
result <- numeric(length(x))
for(i in seq_along(x)) {
  result[i] <- x[i] + y[i]
}

应避免的常见反模式

旧式模式

# Avoid - Old pipe
data %>% function()

# Avoid - Old join syntax
inner_join(x, y, by = c("a" = "b"))

# Avoid - Implicit type conversion
sapply()  # Use map_*() instead

# Avoid - String manipulation in data masking
mutate(data, !!paste0("new_", var) := value)
# Use across() or other approaches instead

性能反模式

# Avoid - Growing objects in loops
result <- c()
for(i in 1:n) {
  result <- c(result, compute(i))  # Slow!
}

# Good - Pre-allocate
result <- vector("list", n)
for(i in 1:n) {
  result[[i]] <- compute(i)
}

# Better - Use purrr
result <- map(1:n, compute)

从旧模式迁移

从 R 基础功能迁移到现代 tidyverse

# Data manipulation
subset(data, condition)          -> filter(data, condition)
data[order(data$x), ]           -> arrange(data, x)
aggregate(x ~ y, data, mean)    -> summarise(data, mean(x), .by = y)

# Functional programming
sapply(x, f)                    -> map(x, f)  # type-stable
lapply(x, f)                    -> map(x, f)

# String manipulation
grepl("pattern", text)          -> str_detect(text, "pattern")
gsub("old", "new", text)        -> str_replace_all(text, "old", "new")
substr(text, 1, 5)              -> str_sub(text, 1, 5)
nchar(text)                     -> str_length(text)
strsplit(text, ",")             -> str_split(text, ",")
paste0(a, b)                    -> str_c(a, b)
tolower(text)                   -> str_to_lower(text)

从旧版 tidyverse 模式迁移到新版模式

# Pipes
data %>% function()             -> data |> function()

# Grouping (dplyr 1.1+)
group_by(data, x) |>
  summarise(mean(y)) |>
  ungroup()                     -> summarise(data, mean(y), .by = x)

# Column selection
across(starts_with("x"))        -> pick(starts_with("x"))  # for selection only

# Joins
by = c("a" = "b")              -> by = join_by(a == b)

# Multi-row summaries
summarise(data, x, .groups = "drop") -> reframe(data, x)

# Data reshaping
gather()/spread()               -> pivot_longer()/pivot_wider()

# String separation (tidyr 1.3+)
separate(col, into = c("a", "b")) -> separate_wider_delim(col, delim = "_", names = c("a", "b"))
extract(col, into = "x", regex)   -> separate_wider_regex(col, patterns = c(x = regex))

已取代的 purrr 函数（purrr 1.0+）

map_dfr(x, f)                   -> map(x, f) |> list_rbind()
map_dfc(x, f)                   -> map(x, f) |> list_cbind()
map2_dfr(x, y, f)               -> map2(x, y, f) |> list_rbind()
pmap_dfr(list, f)               -> pmap(list, f) |> list_rbind()
imap_dfr(x, f)                  -> imap(x, f) |> list_rbind()

# For side effects
walk(x, write_file)             # instead of for loops
walk2(data, paths, write_csv)   # multiple arguments