In cboettig/taxald: A High-Performance Local Taxonomic Database Interface

library(tidyverse)
library(taxadb)

td_create("itis")

bbs <- read_tsv(system.file("extdata/bbs.tsv", package="taxadb"))

name <- bbs$species
authority <- "itis"
db <- td_connect()
match <- "contains"

Strategy 1: SQL-based fuzzy match

name_pattern <- switch(match,
                         starts_with = paste0(name, "%"),
                         contains =  paste0("%", name, "%")
  )

  system.time({
  out <- purrr::map_dfr(name_pattern,
          function(pattern)
            taxa_tbl(authority, "taxonid", db) %>%
            filter(name %like% pattern) %>% collect()
         )
  })
out

Strategy 2: Extract a smaller table

  ## Strategy: extract all potential matches by Genus alone.  assumes first name is a genus name!
  only_genus <- function(name)  stringi::stri_extract_first_words(name)
  id_tbl <- ids(only_genus(name), authority = authority, db = db, collect = FALSE) %>%
    select(name) %>%
    inner_join(select(taxa_tbl(authority, "hierarchy"), id, genus), by = c(name = "genus")) %>%
    select(id) %>% inner_join(taxa_tbl(authority, "taxonid"), by = "id") %>%
    distinct() %>%
    collect()

  name_regex <- switch(match,
                         starts_with = paste0(name, ".*"),
                         contains =  paste0(".*", name, ".*")
  )

  id_tbl <- collect(taxa_tbl(authority, "taxonid"))

  ## Using the genus subset -- a much smaller list of matches -- is this good or bad?
  system.time({
    out2 <- purrr::map_dfr(name_regex, function(pattern)
      filter(id_tbl, grepl(pattern, name))
    )
  })

  ## In memory, even slower!! 
  system.time({
    id_tbl <-  collect(taxa_tbl(authority, "taxonid"))
    out2 <- purrr::map_dfr(name_regex, function(pattern)
      filter(id_tbl, grepl(pattern, name))
    )
  })