threshold_perf: Generate performance metrics across probability thresholds

In topepo/probably: Tools for Post-Processing Predicted Values

Generate performance metrics across probability thresholds

Description

threshold_perf() can take a set of class probability predictions and determine performance characteristics across different values of the probability threshold and any existing groups.

Usage

threshold_perf(.data, ...)

## S3 method for class 'data.frame'
threshold_perf(
  .data,
  truth,
  estimate,
  thresholds = NULL,
  metrics = NULL,
  na_rm = TRUE,
  event_level = "first",
  ...
)

Arguments

.data	A tibble, potentially grouped.
...	Currently unused.
truth	The column identifier for the true two-class results (that is a factor). This should be an unquoted column name.
estimate	The column identifier for the predicted class probabilities (that is a numeric). This should be an unquoted column name.
thresholds	A numeric vector of values for the probability threshold. If unspecified, a series of values between 0.5 and 1.0 are used. Note: if this argument is used, it must be named.
metrics	Either NULL or a yardstick::metric_set() with a list of performance metrics to calculate. The metrics should all be oriented towards hard class predictions (e.g. yardstick::sensitivity(), yardstick::accuracy(), yardstick::recall(), etc.) and not class probabilities. A set of default metrics is used when NULL (see Details below).
na_rm	A single logical: should missing data be removed?
event_level	A single string. Either "first" or "second" to specify which level of truth to consider as the "event".

Details

Note that that the global option yardstick.event_first will be used to determine which level is the event of interest. For more details, see the Relevant level section of yardstick::sens().

The default calculated metrics are:

• yardstick::j_index()

• yardstick::sens()

• yardstick::spec()

• distance = (1 - sens) ^ 2 + (1 - spec) ^ 2

If a custom metric is passed that does not compute sensitivity and specificity, the distance metric is not computed.

Value

A tibble with columns: .threshold, .estimator, .metric, .estimate and any existing groups.

Examples

library(dplyr)
data("segment_logistic")

# Set the threshold to 0.6
# > 0.6 = good
# < 0.6 = poor
threshold_perf(segment_logistic, Class, .pred_good, thresholds = 0.6)

# Set the threshold to multiple values
thresholds <- seq(0.5, 0.9, by = 0.1)

segment_logistic %>%
  threshold_perf(Class, .pred_good, thresholds)

# ---------------------------------------------------------------------------

# It works with grouped data frames as well
# Let's mock some resampled data
resamples <- 5

mock_resamples <- resamples %>%
  replicate(
    expr = sample_n(segment_logistic, 100, replace = TRUE),
    simplify = FALSE
  ) %>%
  bind_rows(.id = "resample")

resampled_threshold_perf <- mock_resamples %>%
  group_by(resample) %>%
  threshold_perf(Class, .pred_good, thresholds)

resampled_threshold_perf

# Average over the resamples
resampled_threshold_perf %>%
  group_by(.metric, .threshold) %>%
  summarise(.estimate = mean(.estimate))

topepo/probably documentation built on Oct. 21, 2024, 3:28 a.m.