yardstick (version 0.0.8)

classification_cost: Costs function for poor classification

Description

classification_cost() calculates the cost of a poor prediction based on user-defined costs. The costs are multiplied by the estimated class probabilities and the mean cost is returned.

Usage

classification_cost(data, ...)
# S3 method for data.frame
classification_cost(
  data,
  truth,
  ...,
  costs = NULL,
  na_rm = TRUE,
  event_level = yardstick_event_level()
)
classification_cost_vec(
  truth,
  estimate,
  costs = NULL,
  na_rm = TRUE,
  event_level = yardstick_event_level(),
  ...
)

Arguments

data

A data.frame containing the truth and estimate columns.

...

A set of unquoted column names or one or more dplyr selector functions to choose which variables contain the class probabilities. If truth is binary, only 1 column should be selected. Otherwise, there should be as many columns as factor levels of truth.

truth

The column identifier for the true class results (that is a factor). This should be an unquoted column name although this argument is passed by expression and supports quasiquotation (you can unquote column names). For _vec() functions, a factor vector.

costs

A data frame with columns "truth", "estimate", and "cost".

"truth" and "estimate" should be character columns containing unique combinations of the levels of the truth factor.

"costs" should be a numeric column representing the cost that should be applied when the "estimate" is predicted, but the true result is "truth".

It is often the case that when "truth" == "estimate", the cost is zero (no penalty for correct predictions).

If any combinations of the levels of truth are missing, their costs are assumed to be zero.

If NULL, equal costs are used, applying a cost of 0 to correct predictions, and a cost of 1 to incorrect predictions.

na_rm

A logical value indicating whether NA values should be stripped before the computation proceeds.

event_level

A single string. Either "first" or "second" to specify which level of truth to consider as the "event". This argument is only applicable when estimator = "binary". The default uses an internal helper that generally defaults to "first", however, if the deprecated global option yardstick.event_first is set, that will be used instead with a warning.

estimate

If truth is binary, a numeric vector of class probabilities corresponding to the "relevant" class. Otherwise, a matrix with as many columns as factor levels of truth. It is assumed that these are in the same order as the levels of truth.

Value

A tibble with columns .metric, .estimator, and .estimate and 1 row of values.

For grouped data frames, the number of rows returned will be the same as the number of groups.

For class_cost_vec(), a single numeric value (or NA).

Details

As an example, suppose that there are three classes: "A", "B", and "C". Suppose there is a truly "A" observation with class probabilities A = 0.3 / B = 0.3 / C = 0.4. Suppose that, when the true result is class "A", the costs for each class were A = 0 / B = 5 / C = 10, penalizing the probability of incorrectly predicting "C" more than predicting "B". The cost for this prediction would be 0.3 * 0 + 0.3 * 5 + 0.4 * 10. This calculation is done for each sample and the individual costs are averaged.

See Also

Other class probability metrics: average_precision(), gain_capture(), mn_log_loss(), pr_auc(), roc_auc(), roc_aunp(), roc_aunu()

Examples

# NOT RUN {
library(dplyr)

# ---------------------------------------------------------------------------
# Two class example
data(two_class_example)

# Assuming `Class1` is our "event", this penalizes false positives heavily
costs1 <- tribble(
  ~truth,   ~estimate, ~cost,
  "Class1", "Class2",  1,
  "Class2", "Class1",  2
)

# Assuming `Class1` is our "event", this penalizes false negatives heavily
costs2 <- tribble(
  ~truth,   ~estimate, ~cost,
  "Class1", "Class2",  2,
  "Class2", "Class1",  1
)

classification_cost(two_class_example, truth, Class1, costs = costs1)

classification_cost(two_class_example, truth, Class1, costs = costs2)

# ---------------------------------------------------------------------------
# Multiclass
data(hpc_cv)

# Define cost matrix from Kuhn and Johnson (2013)
hpc_costs <- tribble(
   ~estimate, ~truth, ~cost,
   "VF",   "VF",     0,
   "VF",    "F",     1,
   "VF",    "M",     5,
   "VF",    "L",    10,
   "F",    "VF",     1,
   "F",     "F",     0,
   "F",     "M",     5,
   "F",     "L",     5,
   "M",    "VF",     1,
   "M",     "F",     1,
   "M",     "M",     0,
   "M",     "L",     1,
   "L",    "VF",     1,
   "L",     "F",     1,
   "L",     "M",     1,
   "L",     "L",     0
)

# You can use the col1:colN tidyselect syntax
hpc_cv %>%
  filter(Resample == "Fold01") %>%
  classification_cost(obs, VF:L, costs = hpc_costs)

# Groups are respected
hpc_cv %>%
  group_by(Resample) %>%
  classification_cost(obs, VF:L, costs = hpc_costs)
# }