hmatch_settle: Sequential hierarchical matching at each hierarchical level,...

In epicentre-msf/hmatch: Tools for Cleaning and Matching Hierarchically-Structured Data

Sequential hierarchical matching at each hierarchical level, settling for the highest resolution match that is possible for each row

Description

Match sets of hierarchical values (e.g. province / county / township) in a raw, messy dataset to corresponding values within a reference dataset, sequentially over each hierarchical level. Specifically, implements hmatch at each successive hierarchical level, starting with only the first level (lowest resolution), then first and second, first second and third, etc.

After the initial matching over all levels, users can optionally use a resolve join to 'settle' for the highest match possible for each row of raw data, even if that match is below the highest-resolution level specified.

Usage

hmatch_settle(
  raw,
  ref,
  pattern,
  pattern_ref = pattern,
  by,
  by_ref = by,
  type = "left",
  allow_gaps = TRUE,
  fuzzy = FALSE,
  fuzzy_method = "osa",
  fuzzy_dist = 1L,
  dict = NULL,
  ref_prefix = "ref_",
  std_fn = string_std,
  ...
)

Arguments

raw	data frame containing hierarchical columns with raw data
ref	data frame containing hierarchical columns with reference data
pattern	regex pattern to match the hierarchical columns in raw Note: hierarchical column names can be matched using either the pattern or by arguments. Or, if neither pattern or by are specified, the hierarchical columns are assumed to be all column names that are common to both raw and ref. See specifying_columns.
pattern_ref	regex pattern to match the hierarchical columns in ref. Defaults to pattern, so only need to specify if the hierarchical columns have different names in raw and ref.
by	vector giving the names of the hierarchical columns in raw
by_ref	vector giving the names of the hierarchical columns in ref. Defaults to by, so only need to specify if the hierarchical columns have different names in raw and ref.
type	type of join ("left", "inner", "anti", "resolve_left", "resolve_inner", or "resolve_anti"). Defaults to "left". See join_types.
allow_gaps	logical indicating whether to allow missing values below the match level, where 'match level' is the highest level with a non-missing value within a given row of raw. Defaults to TRUE.
fuzzy	logical indicating whether to use fuzzy-matching (based on the stringdist package). Defaults to FALSE.
fuzzy_method	if fuzzy = TRUE, the method to use for string distance calculation (see stringdist-metrics). Defaults to "osa".
fuzzy_dist	if fuzzy = TRUE, the maximum string distance to use to classify matches (i.e. a string distance less than or equal to fuzzy_dist will be considered matching). Defaults to 1L.
dict	optional dictionary for recoding values within the hierarchical columns of raw (see dictionary_recoding)
ref_prefix	prefix to add to names of returned columns from ref if they are otherwise identical to names within raw. Defaults to "ref_".
std_fn	function to standardize strings during matching. Defaults to string_std. Set to NULL to omit standardization. See also string_standardization.
...	additional arguments passed to std_fn()

Value

a data frame obtained by matching the hierarchical columns in raw and ref, using the join type specified by argument type (see join_types for more details)

Resolve joins

In a resolve type join with hmatch_settle, rows of raw with multiple matches to ref are resolved to the highest hierarchical level that is non-conflicting among all matches (or no match if there is a conflict at the very first level). E.g.

raw:
⁠1. | United States | <NA> | Jefferson |⁠

Relevant rows from ref:
⁠1. | United States | <NA> | <NA> |⁠
⁠2. | United States | New York | Jefferson |⁠
⁠3. | United States | Pennsylvania | Jefferson |⁠

In a regular join, the single row from raw (above) will match all three rows from ref. However, in a resolve join the multiple matches will be resolved to the first row from ref, because only the first hierarchical level ("United States") is non-conflicting among all possible matches.

Note that there's a distinction between "common" values at a given hierarchical level (i.e. a single unique value in each row) and "non-conflicting" values (i.e. a single unique value or a missing value). E.g.

raw:
⁠1. | United States | New York | New York |⁠

Relevant rows from ref:
⁠1. | United States | <NA> | <NA> |⁠
⁠2. | United States | New York | <NA> |⁠
⁠3. | United States | New York | New York |⁠

In the example above, only the 1st hierarchical level ("United States") is "common" to all matches, but all hierarchical levels are "non-conflicting" (i.e. because row 2 is a hierarchical child of row 1, and row 3 a child of row 2), and so a resolve-type match will be made to the 3rd row in ref.

Examples

data(ne_raw)
data(ne_ref)

# return matches at all levels
hmatch_settle(ne_raw, ne_ref, pattern = "^adm", type = "inner")

# use a resolve join to settle for the best possible match for each row
hmatch_settle(ne_raw, ne_ref, pattern = "^adm", type = "resolve_inner")

epicentre-msf/hmatch documentation built on Nov. 15, 2023, 1:47 a.m.