Compares model predictions against labeled pairs and returns all false-positive and false-negative errors at a given threshold. Useful for understanding which record pairs the model gets wrong.
Arguments
- model
A trained
il_modelobject.- labels
A data frame of labeled pairs with a logical or integer match indicator. Required unless
labels_colis provided.- threshold
A numeric value between 0 and 1 for classifying pairs as matches. Defaults to
0.85.- labels_col
Optional string naming a column in the original data containing ground-truth cluster/entity IDs.
Value
A tibble::tibble() of misclassified pairs with columns unique_id_l,
unique_id_r, match_weight, match_probability, true_label,
and error_type.
Examples
df <- data.frame(
unique_id = 1:20,
first_name = c(
'John', 'Jon', 'Jane', 'Jane', 'Bob',
'Bobby', 'Alice', 'Alicia', 'Tom', 'Thomas',
'John', 'Jon', 'Jane', 'Janet', 'Bob',
'Robert', 'Alice', 'Alison', 'Tom', 'Tomas'
),
surname = c(
'Smith', 'Smith', 'Doe', 'Doe', 'Jones',
'Jones', 'Brown', 'Brown', 'White', 'White',
'Smith', 'Smyth', 'Doe', 'Doe', 'Jones',
'Jones', 'Brown', 'Browne', 'White', 'White'
),
dob = c(
'1990-01-01', '1990-01-01', '1985-06-15', '1985-06-15',
'2000-12-01', '2000-12-01', '1975-03-22', '1975-03-22',
'1988-07-04', '1988-07-04', '1990-01-01', '1990-01-02',
'1985-06-15', '1985-06-16', '2000-12-01', '2000-12-02',
'1975-03-22', '1975-03-23', '1988-07-04', '1988-07-05'
),
city = c(
'London', 'London', 'Paris', 'Paris', 'Berlin',
'Berlin', 'Rome', 'Rome', 'Madrid', 'Madrid',
'London', 'London', 'Paris', 'Paris', 'Berlin',
'Berlin', 'Rome', 'Rome', 'Madrid', 'Madrid'
),
email = c(
'john@example.com', 'jon@example.com', 'jane@example.com',
'jane@example.com', 'bob@example.com', 'bobby@example.com',
'alice@example.com', 'alicia@example.com', 'tom@example.com',
'thomas@example.com', 'john@example.com', 'jon@example.com',
'jane@example.com', 'janet@example.com', 'bob@example.com',
'robert@example.com', 'alice@example.com', 'alison@example.com',
'tom@example.com', 'tomas@example.com'
)
)
con <- DBI::dbConnect(duckdb::duckdb())
spec <- il_spec() |>
il_compare(first_name, cl_jaro_winkler(0.9, 0.7)) |>
il_compare(surname, cl_jaro_winkler(0.9, 0.7)) |>
il_compare(dob, cl_exact()) |>
il_block_on(surname) |>
il_block_on(first_name)
model <- il_model(df, spec = spec, con = con)
model <- il_estimate_u(model)
model <- il_estimate_em(model, block_on(surname))
#> EM trained: first_name and dob | skipped (blocked on): surname
labels <- data.frame(
unique_id_l = c(1L, 1L),
unique_id_r = c(11L, 2L),
is_match = c(1L, 0L)
)
il_errors(model, labels = labels, threshold = 0.85)
#> # A tibble: 1 × 6
#> unique_id_l unique_id_r match_weight match_probability true_label error_type
#> <int> <int> <dbl> <dbl> <lgl> <chr>
#> 1 1 2 8.76 0.998 FALSE false_posit…
DBI::dbDisconnect(con, shutdown = TRUE)