plot_probabilities_ecdf: Plot ECDF for the predicted probabilities

In cvms: Cross-Validation for Model Selection

Plot ECDF for the predicted probabilities

Description

\Sexpr[results=rd, stage=render]{lifecycle::badge("experimental")}

Plots the empirical cumulative distribution function (ECDF) for the probabilities of either the target classes or the predicted classes.

Creates a ggplot2 with the stat_ecdf() geom.

Usage

plot_probabilities_ecdf(
  data,
  target_col,
  probability_cols,
  predicted_class_col = NULL,
  obs_id_col = NULL,
  group_col = NULL,
  probability_of = "target",
  positive = 2,
  theme_fn = ggplot2::theme_minimal,
  color_scale = ggplot2::scale_colour_brewer(palette = "Dark2"),
  apply_facet = length(probability_cols) > 1,
  add_caption = TRUE,
  ecdf_settings = list(),
  facet_settings = list(),
  xlim = c(0, 1)
)

Arguments

data	data.frame with probabilities, target classes and (optional) predicted classes. Can also include observation identifiers and a grouping variable. Example for binary classification: Classifier Observation Probability Target Prediction SVM 1 0.3 cl_1 cl_1 SVM 2 0.7 cl_1 cl_2 NB 1 0.2 cl_2 cl_1 NB 2 0.8 cl_2 cl_2 ... ... ... ... ... Example for multiclass classification: Classifier Observation cl_1 cl_2 cl_3 Target Prediction SVM 1 0.2 0.1 0.7 cl_1 cl_3 SVM 2 0.3 0.5 0.2 cl_1 cl_2 NB 1 0.8 0.1 0.1 cl_2 cl_1 NB 2 0.1 0.6 0.3 cl_3 cl_2 ... ... ... ... ... ... ... As created with the various validation functions in cvms, like cross_validate_fn().
target_col	Name of column with target levels.
probability_cols	Name of columns with predicted probabilities. For binary classification, this should be one column with the probability of the second class (alphabetically). For multiclass classification, this should be one column per class. These probabilities must sum to 1 row-wise.
predicted_class_col	Name of column with predicted classes. This is required when probability_of = "prediction".
obs_id_col	Name of column with observation identifiers for averaging the predicted probabilities per observation before computing the ECDF (when deemed meaningful). When NULL, each row is an observation.
group_col	Name of column with groups. The plot elements are split by these groups and can be identified by their color. E.g. the classifier responsible for the prediction. N.B. With more than 8 groups, the default `color_scale` might run out of colors.
probability_of	Whether to plot the ECDF for the probabilities of the target classes ("target") or the predicted classes ("prediction"). For each row, we extract the probability of either the target class or the predicted class. Both are useful to plot, as they show the behavior of the classifier in a way a confusion matrix doesn't. One classifier might be very certain in its predictions (whether wrong or right), whereas another might be less certain.
positive	TODO
theme_fn	The ggplot2 theme function to apply.
color_scale	ggplot2 color scale object for adding discrete colors to the plot. E.g. the output of ggplot2::scale_colour_brewer() or ggplot2::scale_colour_viridis_d(). N.B. The number of colors in the object's palette should be at least the same as the number of groups in the `group_col` column.
apply_facet	Whether to use ggplot2::facet_wrap(). (Logical) By default, faceting is applied when there are more than one probability column (multiclass).
add_caption	Whether to add a caption explaining the plot. This is dynamically generated and intended as a starting point. (Logical) You can overwrite the text with ggplot2::labs(caption = "...").
ecdf_settings	Named list of arguments for ggplot2::stat_ecdf(). The mapping argument is set separately. Any argument not in the list will use the default value set by cvms. Defaults: list(geom = "smooth", pad = FALSE). Common changes are to set `geom = "step"` and/or `pad = TRUE`.
facet_settings	Named list of arguments for ggplot2::facet_wrap(). The facets argument is set separately. Any argument not in the list will use its default value. Commonly set arguments are nrow and ncol.
xlim	Limits for the x-scale.

Details

TODO

Value

A ggplot2 object with a faceted line plot. TODO

Author(s)

Ludvig Renbo Olsen, [e-mail]

See Also

Other plotting functions: font(), plot_confusion_matrix(), plot_metric_density(), plot_probabilities(), sum_tile_settings()

Examples

# Attach cvms
library(cvms)
library(ggplot2)
library(dplyr)

#
# Multiclass
#

# TODO: Go through and rewrite comments and code!

# Plot probabilities of target classes
# From repeated cross-validation of three classifiers

# plot_probabilities_ecdf(
#   data = predicted.musicians,
#   target_col = "Target",
#   probability_cols = c("A", "B", "C", "D"),
#   predicted_class_col = "Predicted Class",
#   group_col = "Classifier",
#   probability_of = "target"
# )

# Plot probabilities of predicted classes
# From repeated cross-validation of three classifiers

# plot_probabilities_ecdf(
#   data = predicted.musicians,
#   target_col = "Target",
#   probability_cols = c("A", "B", "C", "D"),
#   predicted_class_col = "Predicted Class",
#   group_col = "Classifier",
#   probability_of = "prediction"
# )

#
# Binary
#

# Filter the predicted.musicians dataset
# binom_data <- predicted.musicians %>%
#   dplyr::filter(
#     Target %in% c("A", "B")
#   ) %>%
#   # "B" is the second class alphabetically
#   dplyr::rename(Probability = B) %>%
#   dplyr::mutate(`Predicted Class` = ifelse(
#     Probability > 0.5, "B", "A")) %>%
#   dplyr::select(-dplyr::all_of(c("A","C","D")))

# Plot probabilities of predicted classes
# From repeated cross-validation of three classifiers

# plot_probabilities_ecdf(
#   data = binom_data,
#   target_col = "Target",
#   probability_cols = "Probability",
#   predicted_class_col = "Predicted Class",
#   group_col = "Classifier",
#   probability_of = "target"
# )

# plot_probabilities_ecdf(
#   data = binom_data,
#   target_col = "Target",
#   probability_cols = "Probability",
#   predicted_class_col = "Predicted Class",
#   group_col = "Classifier",
#   probability_of = "prediction",
#   xlim = c(0.5, 1)
# )

cvms documentation built on July 9, 2023, 6:56 p.m.