3 Linking datasets
3.1 Using basic
functionalities of blocking package
The goal of this exercise is to link units from the CIS dataset to the CENSUS dataset.
result1 <- blocking(x = census$txt, y = cis$txt, verbose = 1)
#> ===== creating tokens =====
#> ===== starting search (nnd, x, y: 12671, 12306, t: 1053) =====
#> ===== creating graph =====Distribution of distances for each pair.
Example pairs.
| x | y | block | dist |
|---|---|---|---|
| 1 | 12088 | 8340 | 0.0452079 |
| 2 | 12156 | 8378 | 0.2616496 |
| 3 | 7243 | 5756 | 0.0460410 |
| 3 | 10643 | 5756 | 0.3730645 |
| 4 | 8422 | 6453 | 0.3587636 |
| 6 | 9442 | 7034 | 0.3189948 |
| 6 | 10195 | 7034 | 0.0416667 |
| 7 | 745 | 725 | 0.1633400 |
| 8 | 3072 | 2770 | 0.2578439 |
| 8 | 10717 | 2770 | 0.1235833 |
Let’s take a look at the first pair. Obviously there is a typo in the
pername1 but all the other variables are the same, so it
appears to be a match.
cbind(t(census[1, c(1:7, 9:10)]), t(cis[12088, 1:9]))
#> [,1] [,2]
#> person_id "SW122AB001001" "SW122AB001001"
#> pername1 "GEURGE" "GEORGE"
#> pername2 "HUGHES" "HUGHES"
#> sex "M" "M"
#> dob_day "19" "19"
#> dob_mon "5" "5"
#> dob_year "1942" "1942"
#> enumcap "1 VICTORIA ROAD" "1 VICTORIA ROAD"
#> enumpc "SW122AB" "SW122AB"3.2 Assessing the quality
For some records, we have information about the correct linkage. We can use this information to evaluate our approach.
matches <- merge(x = census[, .(x=1:.N, person_id)],
y = cis[, .(y = 1:.N, person_id)],
by = "person_id")
matches[, block:=1:.N]
head(matches)| person_id | x | y | block |
|---|---|---|---|
| DE03US001003 | 1357 | 10248 | 1 |
| DE03US008001 | 4506 | 2506 | 2 |
| DE03US012002 | 2706 | 12005 | 3 |
| DE03US012003 | 6317 | 11103 | 4 |
| DE03US013003 | 4388 | 10673 | 5 |
| DE03US014003 | 9463 | 11793 | 6 |
So in our example we have 5991 pairs.
result2 <- blocking(x = census$txt, y = cis$txt, verbose = 1,
true_blocks = matches[, .(x, y, block)])
#> ===== creating tokens =====
#> ===== starting search (nnd, x, y: 12671, 12306, t: 1053) =====
#> ===== creating graph =====Let’s see how our approach handled this problem.
result2
#> ========================================================
#> Blocking based on the nnd method.
#> Number of blocks: 8437.
#> Number of columns used for blocking: 1053.
#> Reduction ratio: 0.9999.
#> ========================================================
#> Distribution of the size of the blocks:
#> 2 3 4 5 6 7 8 9
#> 5567 2086 611 142 23 6 1 1
#> ========================================================
#> Evaluation metrics (standard):
#> recall precision fpr fnr accuracy specificity
#> 99.8159 99.5326 0.0001 0.1841 99.9999 99.9999
#> f1_score
#> 99.6740It seems that the default parameters of the NND method result in an
FNR of 0.18%. We can see if decreasing the epsilon
parameter as suggested in the Nearest
Neighbor Descent vignette will help.
ann_control_pars <- controls_ann()
ann_control_pars$nnd$epsilon <- 0.2
result3 <- blocking(x = census$txt, y = cis$txt, verbose = 1,
true_blocks = matches[, .(x, y, block)],
control_ann = ann_control_pars)
#> ===== creating tokens =====
#> ===== starting search (nnd, x, y: 12671, 12306, t: 1053) =====
#> ===== creating graph =====Changing the epsilon search parameter from 0.1 to 0.2
decreased the FNR to 0.07%.
result3
#> ========================================================
#> Blocking based on the nnd method.
#> Number of blocks: 8451.
#> Number of columns used for blocking: 1053.
#> Reduction ratio: 0.9999.
#> ========================================================
#> Distribution of the size of the blocks:
#> 2 3 4 5 6 7 8 9
#> 5592 2079 606 142 25 5 1 1
#> ========================================================
#> Evaluation metrics (standard):
#> recall precision fpr fnr accuracy specificity
#> 99.9332 99.8331 0.0000 0.0668 100.0000 100.0000
#> f1_score
#> 99.8831Finally, compare the NND and HNSW algorithm for this example.
result4 <- blocking(x = census$txt, y = cis$txt, verbose = 1,
true_blocks = matches[, .(x, y, block)],
ann = "hnsw")
#> ===== creating tokens =====
#> ===== starting search (hnsw, x, y: 12671, 12306, t: 1053) =====
#> ===== creating graph =====It seems that the HNSW algorithm also performed with 0.07% FNR.
result4
#> ========================================================
#> Blocking based on the hnsw method.
#> Number of blocks: 8447.
#> Number of columns used for blocking: 1053.
#> Reduction ratio: 0.9999.
#> ========================================================
#> Distribution of the size of the blocks:
#> 2 3 4 5 6 7 8 9
#> 5587 2079 606 142 26 5 1 1
#> ========================================================
#> Evaluation metrics (standard):
#> recall precision fpr fnr accuracy specificity
#> 99.9332 99.8331 0.0000 0.0668 100.0000 100.0000
#> f1_score
#> 99.88313.3 Compare results
Finally, we can compare the results of two ANN algorithms. The overlap between neighbours is given by