"""
Plot a set of ECDF or ERT-ECDF objects into one figure.
The empirical cumulative distribution function (ECDF, see
:mod:`~moptipy.evaluation.ecdf`) is a function that shows the fraction of runs
that were successful in attaining a certain goal objective value over the
time. The combination of ERT and ECDF is discussed in
:mod:`~moptipy.evaluation.ertecdf`.
1. Nikolaus Hansen, Anne Auger, Steffen Finck, Raymond Ros. *Real-Parameter
Black-Box Optimization Benchmarking 2010: Experimental Setup.*
Research Report RR-7215, INRIA. 2010. inria-00462481.
https://hal.inria.fr/inria-00462481/document/
2. Dave Andrew Douglas Tompkins and Holger H. Hoos. UBCSAT: An Implementation
and Experimentation Environment for SLS Algorithms for SAT and MAX-SAT. In
*Revised Selected Papers from the Seventh International Conference on
Theory and Applications of Satisfiability Testing (SAT'04),* May 10-13,
2004, Vancouver, BC, Canada, pages 306-320. Lecture Notes in Computer
Science (LNCS), volume 3542. Berlin, Germany: Springer-Verlag GmbH.
ISBN: 3-540-27829-X. doi: https://doi.org/10.1007/11527695_24.
3. Holger H. Hoos and Thomas Stützle. Evaluating Las Vegas Algorithms -
Pitfalls and Remedies. In Gregory F. Cooper and Serafín Moral, editors,
*Proceedings of the 14th Conference on Uncertainty in Artificial
Intelligence (UAI'98)*, July 24-26, 1998, Madison, WI, USA, pages 238-245.
San Francisco, CA, USA: Morgan Kaufmann Publishers Inc.
ISBN: 1-55860-555-X.
"""
from math import inf, isfinite
from typing import Any, Callable, Final, Iterable, cast
import numpy as np
from matplotlib.artist import Artist # type: ignore
from matplotlib.axes import Axes # type: ignore
from matplotlib.figure import Figure # type: ignore
from pycommons.types import type_error
import moptipy.utils.plot_defaults as pd
import moptipy.utils.plot_utils as pu
from moptipy.evaluation.axis_ranger import AxisRanger
from moptipy.evaluation.base import get_algorithm, sort_key
from moptipy.evaluation.ecdf import Ecdf, get_goal, goal_to_str
from moptipy.evaluation.styler import Styler
from moptipy.utils.lang import Lang
[docs]
def plot_ecdf(ecdfs: Iterable[Ecdf],
figure: Axes | Figure,
x_axis: AxisRanger | Callable[[str], AxisRanger]
= AxisRanger.for_axis,
y_axis: AxisRanger | Callable[[str], AxisRanger]
= AxisRanger.for_axis,
legend: bool = True,
distinct_colors_func: Callable[[int], Any] =
pd.distinct_colors,
distinct_line_dashes_func: Callable[[int], Any] =
pd.distinct_line_dashes,
importance_to_line_width_func: Callable[[int], float] =
pd.importance_to_line_width,
importance_to_alpha_func: Callable[[int], float] =
pd.importance_to_alpha,
importance_to_font_size_func: Callable[[int], float] =
pd.importance_to_font_size,
x_grid: bool = True,
y_grid: bool = True,
x_label: None | str | Callable[[str], str] =
lambda x: x if isinstance(x, str) else x[0],
x_label_inside: bool = True,
y_label: None | str | Callable[[str], str] =
Lang.translate_func("ECDF"),
y_label_inside: bool = True,
algorithm_priority: float = 5.0,
goal_priority: float = 0.333,
algorithm_sort_key: Callable[[str], Any] = lambda x: x,
goal_sort_key: Callable[[str], Any] = lambda x: x,
algorithm_namer: Callable[[str], str] = lambda x: x,
color_algorithms_as_fallback_group: bool = True) -> Axes:
"""
Plot a set of ECDF functions into one chart.
:param ecdfs: the iterable of ECDF functions
:param figure: the figure to plot in
:param x_axis: the x_axis ranger
:param y_axis: the y_axis ranger
:param legend: should we plot the legend?
:param distinct_colors_func: the function returning the palette
:param distinct_line_dashes_func: the function returning the line styles
:param importance_to_line_width_func: the function converting importance
values to line widths
:param importance_to_alpha_func: the function converting importance
values to alphas
:param importance_to_font_size_func: the function converting importance
values to font sizes
:param x_grid: should we have a grid along the x-axis?
:param y_grid: should we have a grid along the y-axis?
:param x_label: a callable returning the label for the x-axis, a label
string, or `None` if no label should be put
:param x_label_inside: put the x-axis label inside the plot (so that
it does not consume additional vertical space)
:param y_label: a callable returning the label for the y-axis, a label
string, or `None` if no label should be put
:param y_label_inside: put the y-axis label inside the plot (so that
it does not consume additional horizontal space)
:param algorithm_priority: the style priority for algorithms
:param goal_priority: the style priority for goal values
:param algorithm_namer: the name function for algorithms receives an
algorithm ID and returns an instance name; default=identity function
:param color_algorithms_as_fallback_group: if only a single group of data
was found, use algorithms as group and put them in the legend
:param algorithm_sort_key: the sort key function for algorithms
:param goal_sort_key: the sort key function for goals
:returns: the axes object to allow you to add further plot elements
"""
# Before doing anything, let's do some type checking on the parameters.
# I want to ensure that this function is called correctly before we begin
# to actually process the data. It is better to fail early than to deliver
# some incorrect results.
if not isinstance(ecdfs, Iterable):
raise type_error(ecdfs, "ecdfs", Iterable)
if not isinstance(figure, Axes | Figure):
raise type_error(figure, "figure", (Axes, Figure))
if not isinstance(legend, bool):
raise type_error(legend, "legend", bool)
if not callable(distinct_colors_func):
raise type_error(
distinct_colors_func, "distinct_colors_func", call=True)
if not callable(distinct_line_dashes_func):
raise type_error(
distinct_line_dashes_func, "distinct_line_dashes_func", call=True)
if not callable(distinct_line_dashes_func):
raise type_error(importance_to_line_width_func,
"importance_to_line_width_func", call=True)
if not callable(importance_to_alpha_func):
raise type_error(
importance_to_alpha_func, "importance_to_alpha_func", call=True)
if not callable(importance_to_font_size_func):
raise type_error(importance_to_font_size_func,
"importance_to_font_size_func", call=True)
if not isinstance(x_grid, bool):
raise type_error(x_grid, "x_grid", bool)
if not isinstance(y_grid, bool):
raise type_error(y_grid, "y_grid", bool)
if not ((x_label is None) or callable(x_label)
or isinstance(x_label, str)):
raise type_error(x_label, "x_label", (str, None), call=True)
if not isinstance(x_label_inside, bool):
raise type_error(x_label_inside, "x_label_inside", bool)
if not ((y_label is None) or callable(y_label)
or isinstance(y_label, str)):
raise type_error(y_label, "y_label", (str, None), call=True)
if not isinstance(y_label_inside, bool):
raise type_error(y_label_inside, "y_label_inside", bool)
if not isinstance(algorithm_priority, float):
raise type_error(algorithm_priority, "algorithm_priority", float)
if not isfinite(algorithm_priority):
raise ValueError(f"algorithm_priority cannot be {algorithm_priority}.")
if not isfinite(goal_priority):
raise ValueError(f"goal_priority cannot be {goal_priority}.")
if not callable(algorithm_namer):
raise type_error(algorithm_namer, "algorithm_namer", call=True)
if not callable(algorithm_sort_key):
raise type_error(algorithm_sort_key, "algorithm_sort_key", call=True)
if not callable(goal_sort_key):
raise type_error(goal_sort_key, "goal_sort_key", call=True)
# First, we try to find groups of data to plot together in the same
# color/style. We distinguish progress objects from statistical runs.
goals: Final[Styler] = Styler(key_func=get_goal,
namer=goal_to_str,
priority=goal_priority,
name_sort_function=goal_sort_key)
algorithms: Final[Styler] = Styler(key_func=get_algorithm,
namer=algorithm_namer,
none_name=Lang.translate("all_algos"),
priority=algorithm_priority,
name_sort_function=algorithm_sort_key)
f_dim: str | None = None
t_dim: str | None = None
source: list[Ecdf] = cast(list[Ecdf], ecdfs) \
if isinstance(ecdfs, list) else list(ecdfs)
del ecdfs
x_labels: set[str] = set()
# First pass: find out the goals and algorithms
for ee in source:
if not isinstance(ee, Ecdf):
raise type_error(ee, "data source", Ecdf)
goals.add(ee)
algorithms.add(ee)
x_labels.add(ee.time_label())
# Validate that we have consistent time and objective units.
if f_dim is None:
f_dim = ee.f_name
elif f_dim != ee.f_name:
raise ValueError(
f"F-units {f_dim} and {ee.f_name} do not fit!")
if t_dim is None:
t_dim = ee.time_unit
elif t_dim != ee.time_unit:
raise ValueError(
f"Time units {t_dim} and {ee.time_unit} do not fit!")
if f_dim is None:
raise ValueError("f_dim cannot be None")
if t_dim is None:
raise ValueError("t_dim cannot be None")
if (source is None) or (len(source) <= 0):
raise ValueError(f"source cannot be {source}.")
# determine the style groups
groups: list[Styler] = []
goals.finalize()
algorithms.finalize()
# pick the right sorting order
sf: Callable[[Ecdf], Any] = sort_key
if (goals.count > 1) and (algorithms.count == 1):
def __x1(r: Ecdf, ssf=goal_sort_key) -> Any:
return ssf(goal_to_str(r.goal_f))
sf = __x1
elif (goals.count == 1) and (algorithms.count > 1):
def __x2(r: Ecdf, ssf=algorithm_sort_key) -> Any:
return ssf(r.algorithm)
sf = __x2
elif (goals.count > 1) and (algorithms.count > 1):
def __x3(r: Ecdf, sgs=goal_sort_key, sas=algorithm_sort_key,
ag=algorithm_priority > goal_priority) -> tuple[Any, Any]:
k1 = sgs(goal_to_str(r.goal_f))
k2 = sas(r.algorithm)
return (k2, k1) if ag else (k1, k2)
sf = __x3
source.sort(key=sf)
def __set_importance(st: Styler) -> None:
none = 1
not_none = 0
none_lw = importance_to_line_width_func(none)
not_none_lw = importance_to_line_width_func(not_none)
st.set_line_width(lambda p: [none_lw if i <= 0 else not_none_lw
for i in range(p)])
none_a = importance_to_alpha_func(none)
not_none_a = importance_to_alpha_func(not_none)
st.set_line_alpha(lambda p: [none_a if i <= 0 else not_none_a
for i in range(p)])
if goals.count > 1:
groups.append(goals)
if algorithms.count > 1:
groups.append(algorithms)
if len(groups) > 0:
groups.sort()
groups[0].set_line_color(distinct_colors_func)
if len(groups) > 1:
groups[1].set_line_dash(distinct_line_dashes_func)
elif color_algorithms_as_fallback_group:
algorithms.set_line_color(distinct_colors_func)
groups.append(algorithms)
# If we only have <= 2 groups, we can mark None and not-None values with
# different importance.
if goals.has_none and (goals.count > 1):
__set_importance(goals)
elif algorithms.has_none and (algorithms.count > 1):
__set_importance(algorithms)
# we will collect all lines to plot in plot_list
plot_list: list[dict] = []
# set up the axis rangers
if callable(x_axis):
x_axis = x_axis(t_dim)
if not isinstance(x_axis, AxisRanger):
raise type_error(x_axis, "x_axis", AxisRanger)
if callable(y_axis):
y_axis = y_axis("ecdf")
if not isinstance(y_axis, AxisRanger):
raise type_error(y_axis, "y_axis", AxisRanger)
# first we collect all ecdf object
max_time: int | float = -inf
max_ecdf: int | float = -inf
max_ecdf_is_at_max_time: bool = False
for ee in source:
style = pd.create_line_style()
for g in groups:
g.add_line_style(ee, style)
x = ee.ecdf[:, 0]
style["x"] = x
x_axis.register_array(x)
y = ee.ecdf[:, 1]
y_axis.register_array(y)
style["y"] = y
plot_list.append(style)
# We need to detect the special case that the maximum time is at
# the maximum ECDF value. In this case, we will later need to extend
# the visible area of the x-axis.
if len(x) < 2:
continue
fy = y[-2]
ft = x[-2]
if isfinite(ft):
if fy >= max_ecdf:
if fy > max_ecdf:
max_ecdf_is_at_max_time = (ft >= max_time)
max_ecdf = fy
else:
max_ecdf_is_at_max_time = max_ecdf_is_at_max_time \
or (ft >= max_time)
elif ft > max_time:
max_ecdf_is_at_max_time = False
max_time = max(max_time, ft)
del source
font_size_0: Final[float] = importance_to_font_size_func(0)
# If the maximum of any ECDF is located directly at the end of the
# x-axis, we need to slightly extend the axis to make it visible.
if max_ecdf_is_at_max_time:
x_axis.pad_detected_range(pad_max=True)
# set up the graphics area
axes: Final[Axes] = pu.get_axes(figure)
axes.tick_params(axis="x", labelsize=font_size_0)
axes.tick_params(axis="y", labelsize=font_size_0)
# draw the grid
if x_grid or y_grid:
grid_lwd = importance_to_line_width_func(-1)
if x_grid:
axes.grid(axis="x", color=pd.GRID_COLOR, linewidth=grid_lwd)
if y_grid:
axes.grid(axis="y", color=pd.GRID_COLOR, linewidth=grid_lwd)
max_x: float = x_axis.get_pinf_replacement()
min_x: float | None = x_axis.get_0_replacement() \
if x_axis.log_scale else None
# plot the lines
for line in plot_list:
x = line["x"]
changed = False
if np.isposinf(x[-1]):
x = x.copy()
x[-1] = max_x
changed = True
if (x[0] <= 0) and (min_x is not None):
if not changed:
changed = True
x = x.copy()
x[0] = min_x
if changed:
line["x"] = x
axes.step(where="post", **line)
del plot_list
x_axis.apply(axes, "x")
y_axis.apply(axes, "y")
if legend:
handles: list[Artist] = []
for g in groups:
g.add_to_legend(handles.append)
g.has_style = False
if algorithms.has_style:
algorithms.add_to_legend(handles.append)
if goals.has_style:
goals.add_to_legend(handles.append)
axes.legend(loc="upper left",
handles=handles,
labelcolor=[art.color if hasattr(art, "color")
else pd.COLOR_BLACK for art in handles],
fontsize=font_size_0)
pu.label_axes(axes=axes,
x_label=" ".join([x_label(x) for x in sorted(x_labels)])
if callable(x_label) else x_label,
x_label_inside=x_label_inside,
x_label_location=1,
y_label=y_label(f_dim) if callable(y_label) else y_label,
y_label_inside=y_label_inside,
y_label_location=0,
font_size=font_size_0)
return axes