yardstick: metric_set – R documentation

Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!

metric_set

Combine metric functions

Description

metric_set() allows you to combine multiple metric functions together into a new function that calculates all of them at once.

Usage

metric_set(...)

Arguments

...

The bare names of the functions to be included in the metric set.

Details

All functions must be either:

Only numeric metrics
A mix of class metrics or class prob metrics

For instance, rmse() can be used with mae() because they are numeric metrics, but not with accuracy() because it is a classification metric. But accuracy() can be used with roc_auc().

The returned metric function will have a different argument list depending on whether numeric metrics or a mix of class/prob metrics were passed in.

# Numeric metric set signature:
fn(
  data,
  truth,
  estimate,
  na_rm = TRUE,
  ...
)

# Class / prob metric set signature:
fn(
  data,
  truth,
  ...,
  estimate,
  estimator =  NULL,
  na_rm = TRUE,
  event_level = yardstick_event_level()
)

When mixing class and class prob metrics, pass in the hard predictions (the factor column) as the named argument estimate, and the soft predictions (the class probability columns) as bare column names or tidyselect selectors to ....

Examples

library(dplyr)

# Multiple regression metrics
multi_metric <- metric_set(rmse, rsq, ccc)

# The returned function has arguments:
# fn(data, truth, estimate, na_rm = TRUE, ...)
multi_metric(solubility_test, truth = solubility, estimate = prediction)

# Groups are respected on the new metric function
class_metrics <- metric_set(accuracy, kap)

hpc_cv %>%
  group_by(Resample) %>%
  class_metrics(obs, estimate = pred)

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

# If you need to set options for certain metrics,
# do so by wrapping the metric and setting the options inside the wrapper,
# passing along truth and estimate as quoted arguments.
# Then add on the function class of the underlying wrapped function,
# and the direction of optimization.
ccc_with_bias <- function(data, truth, estimate, na_rm = TRUE, ...) {
  ccc(
    data = data,
    truth = !! rlang::enquo(truth),
    estimate = !! rlang::enquo(estimate),
    # set bias = TRUE
    bias = TRUE,
    na_rm = na_rm,
    ...
  )
}

# Use `new_numeric_metric()` to formalize this new metric function
ccc_with_bias <- new_numeric_metric(ccc_with_bias, "maximize")

multi_metric2 <- metric_set(rmse, rsq, ccc_with_bias)

multi_metric2(solubility_test, truth = solubility, estimate = prediction)

# ---------------------------------------------------------------------------
# A class probability example:

# Note that, when given class or class prob functions,
# metric_set() returns a function with signature:
# fn(data, truth, ..., estimate)
# to be able to mix class and class prob metrics.

# You must provide the `estimate` column by explicitly naming
# the argument

class_and_probs_metrics <- metric_set(roc_auc, pr_auc, accuracy)

hpc_cv %>%
  group_by(Resample) %>%
  class_and_probs_metrics(obs, VF:L, estimate = pred)

yardstick

Tidy Characterizations of Model Performance

v0.0.8

MIT + file LICENSE

Authors

Max Kuhn [aut], Davis Vaughan [aut, cre], RStudio [cph]

Initial release