nmoo.plotting.performance_indicators

 1"""
 2Performance indicators plotting
 3"""
 4__docformat__ = "google"
 5
 6from typing import Iterable, Optional
 7
 8import seaborn as sns
 9
10from nmoo.benchmark import Benchmark
11
12
13# TODO: replace the benchmark argument by a benchmark_or_results_file_path, and
14# retrieve the relevant benchmark specifications from the csv file.
15def plot_performance_indicators(
16    benchmark: Benchmark,
17    row: Optional[str] = None,
18    *,
19    algorithms: Optional[Iterable[str]] = None,
20    performance_indicators: Optional[Iterable[str]] = None,
21    problems: Optional[Iterable[str]] = None,
22    legend: bool = True,
23    x: str = "n_gen",
24    logscale: bool = False,
25) -> sns.FacetGrid:
26    """
27    Plots all performance indicators in a grid of line plots.
28
29    <center>
30        <img src="https://github.com/altaris/noisy-moo/raw/main/imgs/plot_performance_indicators.png"
31        alt="Example"/>
32    </center>
33
34    The columns of this grid correspond to the performance indicators, whereas
35    the rows can be set to correspond to either `n_run`, `problem` or
36    `algorithm`. For example, if `row="algorithm"` (as is the case above), then
37    each row will correspond to an algorithm, whereas `n_run` and `problem`
38    will be compounded in the line plots. If left to `None`, then `n_run`,
39    `problem` and `algorithm` will all be compounded together.
40
41    Note:
42        If you have the benchmark definition, the `benchmark.csv` file, but do
43        not want to rerun the benchmark, you can use the following trick:
44
45            benchmark = Benchmark(...)                               # Benchmark specification
46            benchmark._results = pd.read_csv(path_to_benchmark_csv)  # Inject results
47            plot_performance_indicators(benchmark, ...)              # Plot
48
49    Args:
50        benckmark: A (ran) benchmark object.
51        row (Optional[str]): See above.
52        algorithms (Optional[Iterable[str]]): List of algorithms to plot,
53            defaults to all.
54        performance_indicators (Optional[Iterable[str]]): List of performance
55            indicators to plot, defaults to all.
56        problems (Optional[Iterable[str]]): List of problems to plot, defaults
57            to all.
58        legend (bool): Wether to display the legend. Defaults to `True`.
59        logscale (bool): Wether to have a logarithmic y scale. Defaults to
60            `False`.
61        x (str): Column for the x axis. Should be among `n_gen` (the default),
62            `n_eval`, or `timedelta`.
63    """
64    if algorithms is None:
65        algorithms = benchmark._algorithms.keys()
66    if performance_indicators is None:
67        performance_indicators = benchmark._performance_indicators
68    if problems is None:
69        problems = benchmark._problems.keys()
70    df = benchmark._results[
71        ["algorithm", "n_eval", "n_gen", "n_run", "problem", "timedelta"]
72        + ["perf_" + pi for pi in performance_indicators]
73    ]
74    df = df[(df.algorithm.isin(algorithms)) & (df.problem.isin(problems))]
75    df.rename(
76        columns={f"perf_{pi}": pi for pi in performance_indicators},
77        inplace=True,
78    )
79    df = df.melt(
80        id_vars=[
81            c for c in df.columns if c not in benchmark._performance_indicators
82        ],
83        var_name="pi",
84    )
85    grid = sns.FacetGrid(df, col="pi", row=row, sharey=False)
86    if row == "algorithm":
87        grid.map_dataframe(sns.lineplot, x=x, y="value", hue="problem")
88    elif row == "problem":
89        grid.map_dataframe(sns.lineplot, x=x, y="value", hue="algorithm")
90    else:
91        grid.map_dataframe(
92            sns.lineplot, x=x, y="value", hue="algorithm", style="problem"
93        )
94    if legend:
95        grid.add_legend()
96    if logscale:
97        grid.set(yscale="log")
98    return grid