Metric that checks for a pattern in the output

metric_contains() makes a metric that is TRUE when the prediction contains pattern. The pattern is fixed when you create the metric: the metric ignores its expected argument, so every row is checked for the same pattern. To compare against a column of the data, use metric_exact_match(), metric_f1() or a custom metric.

Usage

metric_contains(pattern, field = NULL, ignore_case = FALSE, fixed = TRUE)

Arguments

pattern

The text to look for; a regular expression when fixed = FALSE.

field

The output field to search. Give it for outputs with more than one field; with NULL, the whole prediction is searched.

ignore_case

If TRUE, ignore case.

fixed

If TRUE (the default), match pattern as plain text.

Value

A function function(prediction, expected = NULL) returning TRUE or FALSE. field is stored in its "field" attribute.

See also

Other metrics: as_dsprrr_metric(), evaluate(), metric_custom(), metric_exact_match(), metric_f1(), metric_field_match(), metric_threshold(), metric_with_feedback(), metric_with_trace(), vitals_metrics

Examples

cites_source <- metric_contains("[source]", field = "answer")
cites_source(list(answer = "Paris [source]"))
#> [1] TRUE
cites_source(list(answer = "Paris"))
#> [1] FALSE

# The expected value is ignored
cites_source(list(answer = "Paris [source]"), data.frame(answer = "Lyon"))
#> [1] TRUE

# A regular expression
has_number <- metric_contains("[0-9]+", field = "answer", fixed = FALSE)
has_number(list(answer = "The answer is 42"))
#> [1] TRUE