getPairs.R

##### 
# GET PAIRS
####
library(RecordLinkage)

### load data
data(RLdata500)

### make pairs
# man bekommt eine Liste mit:
# "data", "pairs", "frequencies" und "type"
data <- compare.dedup(RLdata500,
                      blockfld = list(1,3,5:7), # focusing of certain constraints
                      strcmp = c(2,3,4),
                      strcmpfun = levenshteinSim)

# overview of data rpairs
summary(data)

### calculate weight
# es gibt zusaetzlich Variable:
?epiWeights

#########
### weight calculation
#########

data <- epiWeights(data)

# Wdata mit den weights
# data is a list
length(data$pairs$id1) # anzahl pairs
length(data$Wdata) # anzahl weights
# overview of data rpairs - weights
head(data$Wdata)


###########
### get pairs
##########
### get paris
?getPairs

# generate pairs of data.frame data
rpairs <- getPairs(data, single.rows=FALSE)
# not number of pairs, if single.row = FALSE, because 1 pairs use 3 rows
length(rpairs$id)
# overview of data rpairs
head(rpairs)


# show all record pairs with weights between 0.5 and 0.6
head(getPairs(data, min.weight=0.5, max.weight=0.6))
tail(getPairs(data, min.weight=0.5, max.weight=0.6))


# show all record pairs with weights between 0.5 and 0.6
getPairs(data, min.weight=0.5, max.weight=0.6)

### make classify
?epiClassify
# es gibt zusätzlich variabel:
# prediction und threshold

# @required data
# @required treshold.upper = 0.5
# treshold.lower = threshold.upper, if not defined
cldata <- epiClassify(data, 0.5)
# possible prediction-levels: Not-linked and linked (No Possible's because only treshold.upper defined)
summary(cldata$prediction)

# threshold.upper = 0.6, threshold.lower = 0.4
cldata <- epiClassify(data, 0.6, 0.4)
summary(cldata$prediction)

# now we can filter , show= "links", "nonlinks", "possible" or "all".
getPairs(cldata, show="links", single.rows=FALSE)