select_n_to_m function

Select matching pairs enforcing one-to-one linkage

## S3 method for class 'cluster_pairs'
select_greedy(
  pairs,
  variable,
  score,
  threshold,
  preselect = NULL,
  id_x = NULL,
  id_y = NULL,
  ...
)

## S3 method for class 'cluster_pairs'
select_n_to_m(
  pairs,
  variable,
  score,
  threshold,
  preselect = NULL,
  id_x = NULL,
  id_y = NULL,
  ...
)

select_greedy(
  pairs,
  variable,
  score,
  threshold,
  preselect = NULL,
  id_x = NULL,
  id_y = NULL,
  ...
)

## S3 method for class 'pairs'
select_greedy(
  pairs,
  variable,
  score,
  threshold,
  preselect = NULL,
  id_x = NULL,
  id_y = NULL,
  x = attr(pairs, "x"),
  y = attr(pairs, "y"),
  inplace = FALSE,
  include_ties = FALSE,
  n = 1L,
  m = 1L,
  ...
)

select_n_to_m(
  pairs,
  variable,
  score,
  threshold,
  preselect = NULL,
  id_x = NULL,
  id_y = NULL,
  ...
)

## S3 method for class 'pairs'
select_n_to_m(
  pairs,
  variable,
  score,
  threshold,
  preselect = NULL,
  id_x = NULL,
  id_y = NULL,
  x = attr(pairs, "x"),
  y = attr(pairs, "y"),
  inplace = FALSE,
  ...
)

Arguments

• pairs: a pairs object, such as generated by pair_blocking
• variable: the name of the new variable to create in pairs. This will be a logical variable with a value of TRUE for the selected pairs.
• score: name of the score/weight variable of the pairs. When not given and attr(pairs, "score") is defined, that is used.
• threshold: the threshold to apply. Pairs with a score above the threshold are selected.
• preselect: a logical variable with the same length as pairs has rows, or the name of such a variable in pairs. Pairs are only selected when preselect is TRUE. This interacts with threshold (pairs have to be selected with both conditions).
• id_x: a integer vector with the same length as the number of rows in pairs, or the name of a column in x. This vector should identify unique objects in x. When not specified it is assumed that each element in x is unique.
• id_y: a integer vector with the same length as the number of rows in pairs, or the name of a column in y. This vector should identify unique objects in y. When not specified it is assumed that each element in y is unique.
• ...: Used to pass additional arguments to methods
• x: data.table with one half of the pairs.
• y: data.table with the other half of the pairs.
• inplace: logical indicating whether pairs should be modified in place. When pairs is large this can be more efficient.
• include_ties: when pairs for a given record have an equal weight, should all pairs be included.
• n: an integer. Each element of x can be linked to at most n elements of y.
• m: an integer. Each element of y can be linked to at most m elements of x.

Returns

Returns the pairs with the variable given by variable added. This is a logical variable indicating which pairs are selected as matches.

Details

Both methods force one-to-one matching. select_greedy uses a greedy algorithm that selects the first pair with the highest weight. select_n_to_m tries to optimise the total weight of all of the selected pairs. In general this will result in a better selection. However, select_n_to_m uses much more memory and is much slower and, therefore, can only be used when the number of possible pairs is not too large.

Note that when include_ties = TRUE the same record can still be selected more than once. In that case the pairs will have an equal weight.

Examples

data("linkexample1", "linkexample2")
pairs <- pair_blocking(linkexample1, linkexample2, "postcode")
pairs <- compare_pairs(pairs, c("lastname", "firstname", "address", "sex"))
model <- problink_em(~ lastname + firstname + address + sex, data = pairs)
pairs <- predict(model, pairs, type = "mpost", add = TRUE, binary = TRUE)

# Select pairs with a mpost > 0.5 and force one-to-one linkage
pairs <- select_n_to_m(pairs, "ntom", "mpost", 0.5)
pairs <- select_greedy(pairs, "greedy", "mpost", 0.5)
table(pairs$ntom, pairs$greedy)

# The same example as above using a cluster;
library(parallel)
cl <- makeCluster(2)

pairs <- cluster_pair_blocking(cl, linkexample1, linkexample2, "postcode")
compare_pairs(pairs, c("lastname", "firstname", "address", "sex"))
model <- problink_em(~ lastname + firstname + address + sex, data = pairs)
predict(model, pairs, type = "mpost", add = TRUE, binary = TRUE)
# Select pairs with a mpost > 0.5 and force one-to-one linkage
# select_n_to_m and select_greedy only work on pairs that are local; 
# therefore we first collect the pairs
select_threshold(pairs, "selected", "mpost", 0.5)
local_pairs <- cluster_collect(pairs, "selected")
local_pairs <- select_n_to_m(local_pairs, "ntom", "mpost", 0.5)
local_pairs <- select_greedy(local_pairs, "greedy", "mpost", 0.5)
table(local_pairs$ntom, local_pairs$greedy)

stopCluster(cl)