gain_curve: Gain curve

Description

gain_curve() constructs the full gain curve and returns a tibble. See gain_capture() for the relevant area under the gain curve. Also see lift_curve() for a closely related concept.

Usage

gain_curve(data, ...)
# S3 method for data.frame
gain_curve(data, truth, ..., na_rm = TRUE)
# S3 method for gain_df
autoplot(object, ...)

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.

na_rm

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

object

The gain_df data frame returned from gain_curve().

Value

A tibble with class gain_df or gain_grouped_df having columns:

.n - The index of the current sample.
.n_events - The index of the current unique sample. Values with repeated estimate values are given identical indices in this column.
.percent_tested - The cumulative percentage of values tested.
.percent_found - The cumulative percentage of true results relative to the total number of true results.

Gain and Lift Curves

The motivation behind cumulative gain and lift charts is as a visual method to determine the effectiveness of a model when compared to the results one might expect without a model. As an example, without a model, if you were to advertise to a random 10% of your customer base, then you might expect to capture 10% of the of the total number of positive responses had you advertised to your entire customer base. Given a model that predicts which customers are more likely to respond, the hope is that you can more accurately target 10% of your customer base and capture >10% of the total number of positive responses.

The calculation to construct gain curves is as follows:

truth and estimate are placed in descending order by the estimate values (estimate here is a single column supplied in ...).
The cumulative number of samples with true results relative to the entire number of true results are found. This is the y-axis in a gain chart.

Multiclass

If a multiclass truth column is provided, a one-vs-all approach will be taken to calculate multiple curves, one per level. In this case, there will be an additional column, .level, identifying the "one" column in the one-vs-all calculation.

Relevant level

There is no common convention on which factor level should automatically be considered the "event" or "positive" result. In yardstick, the default is to use the first level. To change this, a global option called yardstick.event_first is set to TRUE when the package is loaded. This can be changed to FALSE if the last level of the factor is considered the level of interest. For multiclass extensions involving one-vs-all comparisons (such as macro averaging), this option is ignored and the "one" level is always the relevant result.

Details

There is a ggplot2::autoplot() method for quickly visualizing the curve. This works for binary and multiclass output, and also works with grouped data (i.e. from resamples). See the examples.

The greater the area between the gain curve and the baseline, the better the model.

Gain curves are identical to CAP curves (cumulative accuracy profile). See the Engelmann reference for more information on CAP curves.

References

Engelmann, Bernd & Hayden, Evelyn & Tasche, Dirk (2003). "Measuring the Discriminative Power of Rating Systems," Discussion Paper Series 2: Banking and Financial Studies 2003,01, Deutsche Bundesbank.

Examples

Run this code

# NOT RUN {
library(ggplot2)
library(dplyr)

# Two class - a tibble is returned
gain_curve(two_class_example, truth, Class1)

# Use autoplot to visualize
# The top left hand corner of the grey triangle is a "perfect" gain curve
autoplot(gain_curve(two_class_example, truth, Class1))

# Multiclass one-vs-all approach
# One curve per level
hpc_cv %>%
  filter(Resample == "Fold01") %>%
  gain_curve(obs, VF:L) %>%
  autoplot()

# Same as above, but will all of the resamples
# The resample with the minimum (farthest to the left) "perfect" value is
# used to draw the shaded region
hpc_cv %>%
  group_by(Resample) %>%
  gain_curve(obs, VF:L) %>%
  autoplot()

# }

Run the code above in your browser using DataCamp Workspace