Skip to contents

Read a drafting answer

Source: R/ai_parse.R
tfl_ai_parse.Rd

Reads the text a chat model returned for a drafting prompt (tfl_ai_prompt()): the first fenced block (```yaml or ```json) that can be read and starts with the tflspec_ai header; the whole text when no block does. YAML and JSON are both read, whichever the prompt asked for.

Usage

tfl_ai_parse(text, task)

Arguments

text

The answer: a string, or lines of text.

task

One of tfl_ai_tasks(): the task the prompt asked for.

Value

A tfl_ai_answer: a list with task, header (the tflspec_ai mapping), output_id, sheets (toc: a list holding the toc data frame) or design (figure), note, assumptions (character), problems (a data frame where, problem, severity – error or warning), format (yaml or json) and raw (the block's text). sheets / design is NULL when the answer could not be read.

Details

Nothing is repaired: a block YAML cannot read (a {mean} not quoted, a tab) is a problem naming the parser's line, as is a header of another task or version, a column or key the task does not have, a mapping where a value is wanted, or a block that is not closed (an answer cut off). The problems go to tfl_ai_repair_prompt().

  • toc: the rows become a data frame with tfl_read_toc()'s fields as columns, every value text (numbers as text, a list joined with " | ", missing NA).

  • table: each report of outputs is read as tfl_read_output() reads a report file, into outputs (a list of tfl_output by id); a key a report file does not have is a problem. The assumptions (text, or mappings with output_id, where, item, assumption, question) are kept as rows too, in assumption_rows.

  • figure: the design becomes a tfl_fig_design() (the 0.0.12 shape type / style / args is read as one figure layer, and a stats list of before 0.0.24.9074 after the data steps, each with a warning); a note key is kept as the answer's note.

See also

tfl_ai_check() for the checks that need the study's context.

Examples

answer <- c(
  "Here is the TOC:",
  "```yaml",
  "tflspec_ai: {task: toc, version: 1}",
  "toc:",
  "- {output_id: T-14-1-1, type: table, title: Demographics,",
  "   population: Safety Population}",
  "- {output_id: F-14-2-1, type: figure, title: Time to Death}",
  "assumptions:",
  "- No footnotes are given in the SAP.",
  "```")
a <- tfl_ai_parse(answer, "toc")
a$sheets$toc[c("output_id", "type", "title")]
#>   output_id   type         title
#> 1  T-14-1-1  table  Demographics
#> 2  F-14-2-1 figure Time to Death
a$assumptions
#> [1] "No footnotes are given in the SAP."