ROCnGO (version 0.1.0)

summarize_predictor: Summarize classifier performance

Description

Calculates a series of metrics describing global and local classifier performance.

Usage

summarize_predictor(
  data = NULL,
  predictor,
  response,
  ratio,
  threshold,
  .condition = NULL
)

Value

A single row tibble with different predictor with following metrics as columns:

• Area under curve (AUC) as a metric of global performance.

• Partial are under curve (pAUC) as a metric of local performance.

• Indexes derived from pAUC, depending on the selected ratio. Sensitivity indexes will be used for TPR and specificity indexes for FPR.

• Curve shape in the specified region.

Arguments

data

A data.frame or extension (e.g. a tibble) containing values for predictors and response variables.

predictor

A data variable which must be numeric, representing values of a classifier or predictor for each observation.

response

A data variable which must be a factor, integer or character vector representing the prediction outcome on each observation (Gold Standard).

If the variable presents more than two possible outcomes, classes or categories:

• The outcome of interest (the one to be predicted) will remain distinct.

• All other categories will be combined into a single category.

New combined category represents the "absence" of the condition to predict. See .condition for more information.

ratio

Ratio or axis where to apply calculations.

• If "tpr", only points within the specified region of TPR, y axis, will be considered for calculations.

• If "fpr", only points within the specified region of FPR, x axis, will be considered for calculations.

threshold

A number between 0 and 1, both inclusive, which represents the region bound where to calculate partial area under curve.

If ratio = "tpr", it represents lower bound of the TPR region, being its upper limit equal to 1.

If ratio = "fpr", it represents the upper bound of the FPR region, being its lower limit equal to 0.

.condition

A value from response that represents class, category or condition of interest which wants to be predicted.

If NULL, condition of interest will be selected automatically depending on response type.

Once the class of interest is selected, rest of them will be collapsed in a common category, representing the "absence" of the condition to be predicted.

See vignette("selecting-condition") for further information on how automatic selection is performed and details on selecting the condition of interest.

Examples

# Summarize Sepal.Width as a classifier of setosa species
# and local performance in TPR (0.9, 1)
summarize_predictor(
 data = iris,
 predictor = Sepal.Width,
 response = Species,
 ratio = "tpr",
 threshold = 0.9
)
# Summarize Sepal.Width as a classifier of setosa species
# and local performance in FPR (0, 0.1)
summarize_predictor(
 data = iris,
 predictor = Sepal.Width,
 response = Species,
 ratio = "fpr",
 threshold = 0.1
)