在循环中匹配不同长度的data.tables

Question

Building on a previous question ( R function to failback in a left_join? ), I have 24 different data tables that each use an industry classification system called NAICS and I want to find the best industry match in each table for a given list of industries.

The industry codes get less detailed as they get shorter, so if there isn't an exact match, I want a slightly shorter version of the target. For example, using classification code 311111 as the target:

1. One table may have an exact match: 311111
2. One table may have one level less detailed: 31111
3. One table may only have a much less detailed match: 31

Current approach (see below for code): Loop through all of the tables and then loop through each code length (311111, 31111, 3111, 311, 31, 3) and try to find a match in that table.

My problem:

How do I adjust the code so that multiple instances of a match don't create an error (as in Supplied 261022 items to be assigned to 360 items of column 'match' )? Some data is time series data, so the same industry code will be listed with 100 or more observations. Some data is cross sectional, so the industry codes only appear once.

Full code for context, but question refers to Step 4 :

library(data.table)

# Step 1: Load Table Data -------------------------------------------------
v_tablenames <- c("t_naics17index", "t_naics17def", "t_naics17cross", "t_naics17tree", 
                  "t_naics17isic4cross", "t_ios_2012", "t_iou_2012", "t_regdata6dig_2017", 
                  "t_brdis_2015", "t_mrkcon_2012", "t_matkind_2012", "t_ppiprice", 
                  "t_eximprice", "t_oes", "t_ces", "t_cps", "t_fed", "t_asm", "t_vps", 
                  "t_cbp", "t_exports", "t_imports", "t_expartner", "t_impartner")

for(tablename in v_tablenames){
  assign(tablename, readRDS(paste0("DataStore/", tablename, ".rds")))
}

# Step 2: Turn all of the tibbles into data.tables ------------------------
# Data wrangling done in the tidyverse; tibbles converted to data.tables
l_tables <- list(t_naics17index, t_naics17def, t_naics17cross, t_naics17tree, 
                 t_naics17isic4cross, t_ios_2012, t_iou_2012, t_regdata6dig_2017,
                 t_brdis_2015, t_mrkcon_2012, t_matkind_2012, t_ppiprice, 
                 t_eximprice, t_oes, t_ces, t_cps, t_fed, t_asm, t_vps, 
                 t_cbp, t_exports, t_imports, t_expartner, t_impartner)

lapply(l_tables, setDT)

# Step 3: Build Master Lookup Table ---------------------------------------
# Subset of classification codes I care about falls between 3----- and 4-----; pulled from t_naics17index, which has a complete list of codes
t_match <- unique(t_naics17index[NAICS17 >= "300000" & NAICS17 < "400000", c(1)])

# Step 4: Connect Data Tables ---------------------------------------------
code_len_count <- rev(seq_len(max(nchar(t_match$NAICS17))))

for (tablename in v_tablenames){
  t_match[, match := NA_character_]
  for (i in code_len_count){
    t_match[is.na(match), target := substr(NAICS17, 1, i)]
    t_match[is.na(match), match := get(tablename)[.SD, on=.(NAICS17 = target), mget("NAICS17")][]]
  }
  setnames(t_match, "match", paste0("m_", tablename))
}

Data examples:

# Table of target industry codes
t_match <- structure(list(NAICS17 = c("311111", "311119", "311211", "311212", 
"311213", "311221", "311224", "311225", "311230", "311313")), row.names = c(NA, 
-10L), class = "data.frame")

# NAICS17 column is unique:
t_naics17tree <- structure(list(NAICS17 = c("31-33", "311", "3111", "31111", "311111", 
"311119", "3112", "31121", "311211", "311212"), NAICS17Title = c("Manufacturing", 
"Food Manufacturing", "Animal Food Manufacturing", "Animal Food Manufacturing", 
"Dog and Cat Food Manufacturing", "Other Animal Food Manufacturing", 
"Grain and Oilseed Milling", "Flour Milling and Malt Manufacturing", 
"Flour Milling", "Rice Milling")), row.names = c(NA, 10L), class = "data.frame")

# NAICS17 column is NOT unique:
t_ppiprice <- structure(list(NAICS17 = c("311---", "311---", "311---", "311---", 
"311---", "311---", "311---", "311---", "311---", "311---"), 
    seriesID = c("PCU311---311---", "PCU311---311---", "PCU311---311---", 
    "PCU311---311---", "PCU311---311---", "PCU311---311---", 
    "PCU311---311---", "PCU311---311---", "PCU311---311---", 
    "PCU311---311---"), date = structure(c(17956, 17928, 17897, 
    17866, 17836, 17805, 17775, 17744, 17713, 17683), class = "Date"), 
    value = c(199.2, 198.9, 198.3, 197.9, 197.2, 197.4, 197.1, 
    197.7, 198.8, 200.2)), class = "data.frame", row.names = c(NA, 
-10L))

Answer 1

For posterity, I figured it out...

for (tablename in v_tablenames){
  t_match[, match := NA_character_]
  for (i in code_len_count){
    t_match[is.na(match), target := substr(NAICS17, 1, i)]
    t_match[is.na(match), match := get(paste0("t_", tablename))[.SD, on=.(NAICS17 = target), mult = "first", mget("x.NAICS17")][]]
  }
  setnames(t_match, "match", paste0("m_", tablename))
}

Adding get() around tablename allows the loop to reference the variables and the variable names.

Adding mult = "first" allows the join to only take the first match

Thanks for the help @Cole!