[display]

The figures produced for each outcome. This page describes the figures of a classifier. The regressor has its own defaults and figures, which are described on the regression page. The table is required when run_mode is "audit" or "eval", the modes that produce figures, and it can be left out otherwise.

[display.defaults]
n_bootstraps = 1000
save = true
show = false
n_bins = 10
dist_n_bins = 25
dist_yscale = "log"
strategy = "uniform"

[display.overrides.calibration]
n_bootstraps = 10

[display.outcome_overrides.ANY_COMP.reliability]
strategy = "quantile"

[display.theme]
dpi = 300

Key	Type	Default	Description
defaults	table	all defaults	Settings used by every figure.
overrides	table	none	Settings for one type of figure, for all outcomes.
outcome_overrides	table	none	Settings for one type of figure, for one outcome.
theme	table	none	Fonts, colours and resolution of the figures.

The figures of each outcome are saved in the plots/<outcome>/ folder of the run, and the subgroup heatmaps in plots/. For a classifier these are:

Figure	File	Plot type
ROC curve	<outcome>_roc_curve.png	roc
Precision-recall curve	<outcome>_pr_curve.png	precision_recall
Reliability diagram (calibration curve)	<outcome>_reliability_diagram.png	reliability
Histogram of the predicted probabilities	<outcome>_data_distribution.png	data_distribution
Decision curve analysis	<outcome>_dca_curve.png	dca
Subgroup heatmap, one per metric	<metric>_strata_heatmap.png	strata_heatmap

The heatmaps are only drawn when [workflow.evaluation.fairness] is defined. Each has one row for the whole test set and one for every subgroup, and one column for every outcome. The text is the value of the metric, and the colour is how far the value is from the one of the whole test set.

[display.defaults]

Settings that apply to every figure.

[display.defaults]
n_bootstraps = 1000
save = true
show = false
n_bins = 10
dist_n_bins = 25
dist_yscale = "log"
strategy = "uniform"

Key	Type	Default	Description
n_bootstraps	integer	1000	Number of bootstrap resamples for the 95% confidence bands of the curves. With 0, the bands are not drawn.
save	boolean	true	Whether the figure is saved in the run folder.
show	boolean	false	Whether the figure is displayed in a window.
n_bins	integer	10	Number of bins of the reliability diagram.
strategy	string	"uniform"	How the reliability diagram is computed: "uniform" (bins of equal width), "quantile" (bins with the same number of samples) or "spline" (a smooth curve fitted with splines).
cv_splines	integer	3	Number of internal folds of the spline fit, 2 or more. Only used when strategy is "spline".
dist_n_bins	integer	10	Number of bins of the histogram of the predicted probabilities. It is also used for the histogram drawn under the reliability diagram.
dist_yscale	string	"linear"	Scale of the y-axis of these histograms: "linear", "log", "symlog" or "logit".
n_jobs	integer	1	Number of parallel jobs for the bootstrap of the figures. It is independent of workflow.n_jobs.

Note

The histograms have their own dist_n_bins and dist_yscale. n_bins only applies to the reliability diagram.

[display.overrides] and [display.outcome_overrides]

The defaults can be changed for one type of figure with overrides, and for one type of figure of one outcome with outcome_overrides. Both use the same keys as defaults, and only the keys that are written are changed.

# Changes the calibration figures of all outcomes
[display.overrides.calibration]
n_bootstraps = 10

[display.overrides.roc]
n_bootstraps = 500

# Changes the reliability diagram of ANY_COMP only
[display.outcome_overrides.ANY_COMP.reliability]
n_bootstraps = 50
strategy = "quantile"

For a given figure, the settings are combined in this order, and the last one wins:

1. [display.defaults]
2. [display.overrides.<plot type>]
3. [display.outcome_overrides.<outcome>.<plot type>]
4. The arguments given when a plot method is called in Python.

The plot type is the name of the figure in the table below. Each type also has aliases, which can be used in any of the tables, and the different names of a type are combined.

Plot type	Aliases	Keys that have an effect
roc	roc_curve	n_bootstraps, n_jobs, save, show
precision_recall	pr, pr_curve	n_bootstraps, n_jobs, save, show
reliability	calibration, reliability_diagram	n_bins, strategy, cv_splines, dist_n_bins, dist_yscale, n_bootstraps, n_jobs, save, show
data_distribution	distribution, dist	dist_n_bins, dist_yscale, save, show
dca	dca_curve	save, show
strata_heatmap	heatmap	save, show

The name of the outcome in outcome_overrides must be in data.outcomes. A plot type that is not in the table is rejected. A key that has no effect for a plot type, such as n_bins for the ROC curve, is accepted and ignored.

[display.theme]

The look of all the figures.

[display.theme]
primary_color = "#2D90D8"
ci_alpha = 0.3
dpi = 300
font_family = "sans-serif"
show_grid = false

Key	Type	Default	Description
primary_color	string	"#2D90D8"	Colour of the curves and bars. The heatmaps have their own colour scale.
palette	list of strings	five colours	Reserved for figures that draw several series. Currently not used.
ci_alpha	float	0.3	Opacity of the confidence bands, between 0 and 1.
linewidth	float	2.0	Width of the curves.
dpi	integer	300	Resolution of the saved images.
font_family	string	"sans-serif"	Font of the titles, labels and ticks.
title_fontsize	integer	12	Size of the titles.
label_fontsize	integer	10	Size of the axis labels.
show_spines	boolean	false	Whether the top and right borders of the axes are drawn.
show_grid	boolean	false	Whether a grid is drawn.

A key that is not in the table is rejected. Colours can be written as hexadecimal codes or as any colour name that Matplotlib accepts.