Skip to contents

Ask System One questions about every row of a data frame

Source: R/system_one_df.R
system_one_df.Rd

Sends one request per row, with every question batched into that request, and performs the requests in parallel with httr2::req_perform_parallel(). Rate-limited and failed requests are retried as described in typesafer_error.

Usage

system_one_df(
  .data,
  state,
  ...,
  model = ts_default_model(),
  probs = FALSE,
  include_usage = FALSE,
  max_active = 10,
  on_error = c("stop", "continue"),
  max_tries = 3,
  timeout = ts_timeout(),
  api_key = ts_api_key()
)

Arguments

.data

A data frame.

state

<data-masking> An expression evaluated in .data giving the state for each row: usually a character column (state = text), an expression such as paste(subject, body, sep = "\n\n"), or a list-column holding one structured state (a list) per row.

...

Named questions created with ts_noul(), ts_choice(), or ts_score(). The names identify the answers and aren't shown to the model. You can also splice a named list of questions with !!!.

model

The model to use. Defaults to ts_default_model(), which is "jev-latest", TypeSafe's flagship model, unless the TYPESAFE_DEFAULT_MODEL environment variable says otherwise. Pin a versioned ID such as "jev-1.13.0" to keep answers stable. See ts_models().

probs

If TRUE, also add a <name>_probs list-column for each choice and score question holding its full probability distribution.

include_usage

If TRUE, also add input_tokens and output_tokens columns with each row's token usage (NA for failed rows).

max_active

Maximum number of requests in flight at once.

on_error

What to do when a row's request fails after retries:

  • "stop" (default): stop sending requests and raise the error for the first failed row (see typesafer_error); the condition has a row field.

  • "continue": finish every row, fill the answers of failed rows with NA, add a .error list-column holding each row's error (or NULL), and warn with a count of failed rows.

max_tries

Maximum number of attempts per request, including the first. Requests that fail with 408, 429, 5xx, or a connection failure are retried with exponential backoff, within a 30 second retry budget. Use 1 to disable retries.

timeout

Timeout for each attempt, in seconds. Defaults to the typesafer.timeout option, or 10.

api_key

API key; defaults to ts_api_key().

Value

.data as a tibble, with columns added for each question name:

  • noul: name (probability of yes).

  • choice: name (the chosen option) and name_confidence, plus name_probs (named double vectors) when probs = TRUE.

  • score: name (0-based expected level) and name_confidence, plus name_probs (double vectors ordered by level) when probs = TRUE.

Plus input_tokens and output_tokens when include_usage = TRUE, and .error when on_error = "continue".

See also

system_one() for a single request.

Examples

if (FALSE) { # \dontrun{
tickets <- tibble::tibble(
  id = 1:3,
  text = c(
    "I was charged twice. Please help ASAP.",
    "How do I export my data to CSV?",
    "Your app crashed and I lost an hour of work!!"
  )
)
system_one_df(
  tickets,
  state = text,
  billing = ts_noul("Is this about billing?"),
  tone = ts_choice("What is the tone?", c("calm", "angry")),
  urgency = ts_score("How urgent is this?", c("low", "medium", "high"))
)
} # }