Skip to contents

The study facts a drafting prompt carries

Source: R/ai_context.R
tfl_ai_context.Rd

One constructor for what tfl_ai_prompt() tells the model about the study and the report, and what tfl_ai_check() checks the answer against, so a caller cannot pass the wrong shape and a prompt cannot read a field that is not there. Each task reads its own fields; a field another task reads is an error. Every field is optional but output_id for a figure.

Usage

tfl_ai_context(
  task,
  study = NULL,
  output_id = NULL,
  titles = NULL,
  population = NULL,
  datasets = NULL,
  design = NULL,
  templates = NULL,
  toc_fields = NULL,
  reports = NULL,
  language = NULL,
  outputs = NULL,
  shell = NULL,
  study_dir = NULL,
  examples = NULL
)

Arguments

task

One of tfl_ai_tasks(): "toc", "figure" or "table".

study

The study: a named list or character vector with study_id and title (either may be missing).

output_id

figure: the report's id (required).

titles

figure: the report's title lines, as the TOC has them.

population

figure: its analysis set: one label, or a named list or vector with id, label and where (the condition, such as FASFL == "Y").

datasets

figure, toc and table: the datasets and variables the answer may name. A tfl_ai_adam_summary() (its variables); or a data frame, one row a variable, with dataset and variable and optionally label, kind, levels (| between the values) and dataset_label – tfl_adam_metadata()'s result is one (its codelist and source are not sent); or the ADaM datasets themselves, a named list of data frames, of which only the names, the labels (the label attribute) and the kinds are taken.

design

figure: the current design (a tfl_fig_design()), when there is one: the prompt shows it and asks for it whole, changed.

templates

figure: the templates the answer may start from (names of tfl_fig_templates()); NULL for all.

toc_fields

toc: the columns the answer fills, among output_id (always), type, title, population, footnote, program, file, note, section, datasets, label (tfl_read_toc()'s fields); NULL for output_id, type, title, population, footnote, section, datasets, label.

reports

toc: the reports already in the TOC: their ids, or a data frame with output_id and optionally title.

language

The language the model writes its assumptions in, such as "Japanese"; NULL: the language of the user's message. The prompt itself is English.

outputs

table: the reports to draft (5 to 10: one batch of tfl_ai_batches()), as the TOC numbers them.

shell

table: the TOC and shells as text – a tfl_ai_read_shell() result (read again for outputs alone when it holds the whole workbook) or a string; NULL when the workbook is attached to the chat instead.

study_dir

table: the study folder (tfl_ai_study_init()): its spec/data.yml (the analysis sets the reports name) and spec/defaults.yml (what every report extends) go into the prompt.

examples

table: reviewed report files of this study (spec/outputs/<id>.yml paths) that the new reports follow – a family's first table, corrected by hand, for the rest of the family.

Value

A tfl_ai_context: a list with task and the fields given.

Details

Nothing here is sent anywhere: the context becomes text in the prompt, which the user reads before copying it. Only names and labels are carried, never the data's values: a dataset's levels (a variable's values, such as its PARAMCDs) are there only when the caller gives them.

See also

tfl_ai_prompt().

Examples

tfl_ai_context("toc", study = list(
  study_id = "XYZ-ONC-01", title = "Drug X versus control in advanced solid tumours"
))
#> <tfl_ai_context: toc>
#> ## The study
#> 
#> XYZ-ONC-01: Drug X versus control in advanced solid tumours
#> 
#> ## The TOC's columns
#> 
#> Fill these columns (the schema describes them): `output_id`, `type`, `title`, `population`, `footnote`, `section`, `datasets`, `label`.
#> 
#> ## Reports already in the TOC
#> 
#> None: the TOC is empty.

adam <- tfl_example_adam()
tfl_ai_context("figure", output_id = "F-14-2-1",
               titles = "Kaplan-Meier Plot of Time to Death",
               population = list(id = "FAS", label = "Full Analysis Set",
                                 where = "FASFL == \"Y\""),
               datasets = adam["ADTTE"])
#> <tfl_ai_context: figure>
#> ## The report
#> 
#> - Report: F-14-2-1
#> - Titles: Kaplan-Meier Plot of Time to Death
#> - Analysis set: FAS, Full Analysis Set (`FASFL == "Y"`)
#> 
#> ## The datasets
#> 
#> Name only these datasets and variables.
#> 
#> ### ADTTE
#> 
#> | variable | label | kind |
#> |---|---|---|
#> | USUBJID |  | character |
#> | TRT01P |  | character |
#> | TRT01PN |  | numeric |
#> | FASFL | Full Analysis Set Population Flag | character |
#> | PARAMCD |  | character |
#> | PARAM |  | character |
#> | AVAL | Analysis Value | numeric |
#> | CNSR | Censor | numeric |
#> | SEX |  | character |
#> | AGEGR1 |  | character |
#> 
#> ## The current design
#> 
#> There is none yet: start from the template that fits best.