Answer and response types

system_one() returns a ts_response holding one answer per question:

• ts_answer_noul: @prob, the probability (0 to 1) that the answer is yes. Nouls have no separate confidence.

• ts_answer_choice: @choice, the highest-probability option; @probabilities, a named double vector with one probability per option; and @confidence.

• ts_answer_score: @score, the probability-weighted level, on the API's 0-based scale (with levels c("low", "medium", "high"), 1.8 is close to "high"); @probabilities, a double vector ordered by level, so element i is the probability of level i - 1; @legend, the level descriptions in the same order; and @confidence.

Confidence runs from 0 to 1 and summarizes how concentrated the probability distribution is. See https://docs.typesafe.ai/confidence.

A ts_response also keeps the parsed response body in @json, so you can reach fields typesafer doesn't model yet (including answers of question types it doesn't recognize). Its structure follows the API and may change.

Usage

ts_answer_noul(prob = numeric(0))

ts_answer_choice(
  choice = character(0),
  probabilities = numeric(0),
  confidence = numeric(0)
)

ts_answer_score(
  score = numeric(0),
  probabilities = numeric(0),
  legend = list(),
  confidence = numeric(0)
)

ts_response(
  answers = list(),
  model = character(0),
  usage = integer(0),
  json = list()
)

Arguments

prob, choice, probabilities, score, legend, confidence

Answer fields; see Description.

answers

A named list of answers.

model

The model that answered, e.g. "jev-1.13.0".

usage

A named integer vector with input_tokens and output_tokens.

json

The parsed response body, as a list.

Value

An S7 object.

Examples

ts_response(
  answers = list(
    is_urgent = ts_answer_noul(prob = 0.95),
    department = ts_answer_choice(
      choice = "billing",
      probabilities = c(billing = 0.88, technical = 0.12, sales = 0),
      confidence = 0.81
    )
  ),
  model = "jev-1.13.0",
  usage = c(input_tokens = 318L, output_tokens = 34L)
)
#> <ts_response> jev-1.13.0 | 318 input / 34 output tokens
#>   is_urgent  noul   0.95
#>   department choice "billing" (confidence 0.81)