Parsers for model responses

A parser is a function that turns the raw text response of the backend into a single value: a number, a category label, or NA if the response cannot be interpreted. Parse failures are recorded in the run log and reported as part of the validation (a response that cannot be parsed is a failed measurement).

Usage

parse_identity(text)

parse_number(decimal = ".")

parse_choice(levels, aliases = NULL, whole_word = FALSE)

parse_json_field(field)

Arguments

text

The raw response text (argument of the returned parser and of parse_identity()).

decimal

Decimal mark for parse_number(). Both ASCII minus and the Unicode minus sign U+2212 are accepted.

levels

For parse_choice(): the admissible labels, in the order that defines the ordinal scale. Matching is case-insensitive; the longest label occurring in the response wins, so that "likely pathogenic" is not matched as "pathogenic".

aliases

Optional named list mapping additional strings to labels, e.g. list("LP" = "Likely pathogenic").

whole_word

If TRUE, split at non-alphanumeric characters and match complete word sequences, longest first. Matched words are consumed; different labels in one response give NA. Repeated matches of the same label are allowed. The default retains the substring parser.

field

For parse_json_field(): the name of the field to extract from a JSON object in the response (the first {...} block is used). The complete first object must be valid JSON; only its top-level field is read. Strings are decoded and numbers returned as character values. Missing fields, malformed input and non-scalar field values return NA. The factory requires jsonlite; without it the diagnostic is: "Package jsonlite is required for parse_json_field(); install it to use this parser."

Value

A function with signature function(text) returning a length-one value.

Examples

parse_number()("The IC50 is approximately 12.5 nM.")
#> [1] 12.5
parse_choice(c("Benign", "Likely benign", "VUS", "Likely pathogenic", "Pathogenic"))(
  "Classification: likely pathogenic (PS1, PM2)")
#> [1] "Likely pathogenic"
parse_json_field("class")('Result: {"class": "VUS", "criteria": ["PM2"]}')
#> [1] "VUS"