Merge pull request #9 from OpenSourceEconomics/figures

src/tranquilo_dev/config.py

@@ -71,6 +71,32 @@ def _n_evals_10(*args, **kwargs):  # noqa: U100
     return 10
 
 
+LABELS = {
+    # Tranquilo labels
+    "tranquilo": "Tranquilo-Scalar",
+    "tranquilo_default": "Tranquilo-Scalar",
+    "tranquilo_ls": "Tranquilo-LS",
+    "tranquilo_ls_default": "Tranquilo-LS",
+    "tranquilo_ls_parallel_2": "Tranquilo-LS (2 cores)",
+    "tranquilo_ls_parallel_4": "Tranquilo-LS (4 cores)",
+    "tranquilo_ls_parallel_8": "Tranquilo-LS (8 cores)",
+    "tranquilo_experimental": "Tranquilo-Scalar-Experimental",
+    "tranquilo_ls_experimental": "Tranquilo-LS-Experimental",
+    # DFO-LS labels
+    "dfols": "DFO-LS",
+    "dfols_noisy_3": "DFO-LS (3 evals)",
+    "dfols_noisy_5": "DFO-LS (5 evals)",
+    "dfols_noisy_10": "DFO-LS (10 evals)",
+    # Other labels
+    "nag_bobyqa": "NAG-BOBYQA",
+    "nlopt_bobyqa": "NlOpt-BOBYQA",
+    "nlopt_neldermead": "NlOpt-Nelder-Mead",
+    "scipy_neldermead": "SciPy-Nelder-Mead",
+    "tao_pounders": "TAO-Pounders",
+    "pounders": "Pounders",
+}
+
+
 COMPETITION = {
     "nlopt_bobyqa": {"algorithm": "nlopt_bobyqa"},
     "nag_bobyqa": {"algorithm": "nag_pybobyqa"},

src/tranquilo_dev/plotting/benchmark_plotting_functions.py

@@ -0,0 +1,269 @@
+"""Plotting function that create the publication ready figures.
+
+Most of the plotting behavior can be configured by changing some global dictionaries:
+
+- `AXIS_LABELS`: Contains the axis labels for each plot type.
+- `X_RANGE`: Contains the x-axis range given each benchmark.
+- `COLORS`: Contains the colors for each line given each benchmark.
+- `LINE_WIDTH_UPDATES`: Contains the line width for each line given each benchmark.
+- `LEGEND_LABEL_ORDER`: Contains the order of the legend labels given each benchmark.
+
+"""
+import itertools
+
+import matplotlib
+import matplotlib.pyplot as plt
+from tranquilo_dev.config import LABELS
+
+
+# ======================================================================================
+# Plot Config
+# ======================================================================================
+
+FIGURE_WIDTH_IN_CM = 14.69785
+FIGURE_HEIGHT_IN_CM = 10.0
+
+AXIS_LABELS = {
+    "profile_plot": {
+        "xlabel": "Computational budget (normalized)",
+        "ylabel": "Share of solved problems",
+    },
+    "deviation_plot": {
+        "xlabel": "Computational budget",
+        "ylabel": "Average distance to optimum (normalized)",
+    },
+}
+
+X_RANGE = {
+    "profile_plot": {
+        "scalar_benchmark": (1, 50),
+        "ls_benchmark": (1, 40),
+        "parallel_benchmark": (1,),
+        "noisy_benchmark": (1,),
+        "scalar_vs_ls_benchmark": (1, 50),
+    },
+    "deviation_plot": {
+        "scalar_benchmark": (0, 300),
+        "ls_benchmark": (0, 400),
+        "parallel_benchmark": (0, 50),
+        "noisy_benchmark": (0,),
+        "scalar_vs_ls_benchmark": (0, 500),
+    },
+}
+
+# Colors
+# ======================================================================================
+
+DARK_GRAY = "#534a46"
+
+TABLEAU_10_COLORS = {
+    "blue": "#5778a4",
+    "green": "#6a9f58",
+    "orange": "#e49444",
+    "red": "#d1615d",
+    "teal": "#85b6b2",
+    "yellow": "#e7ca60",
+    "purple": "#a87c9f",
+    "pink": "#f1a2a9",
+    "brown": "#967662",
+    "gray": "#b8b0ac",
+    # Additional shades. (Taken by plugging in the color code of blue and green at
+    # https://www.w3schools.com/colors/colors_picker.asp, and selecting a lighter /
+    # darker color given some percentage.)
+    "green-25": "#b7d2ad",
+    "green-50": "#6ea45b",
+    "green-75": "#37522d",
+    "blue-75": "#abbed4",
+    "blue-55": "#688ab1",
+    "blue-35": "#3d5776",
+    "blue-15": "#1a2532",
+}
+
+
+BASE_COLORS = {
+    "Tranquilo-Scalar": TABLEAU_10_COLORS["teal"],
+    "Tranquilo-LS": TABLEAU_10_COLORS["blue"],
+    "DFO-LS": TABLEAU_10_COLORS["green"],
+    "NAG-BOBYQA": TABLEAU_10_COLORS["orange"],
+    "NlOpt-BOBYQA": TABLEAU_10_COLORS["red"],
+    "NlOpt-Nelder-Mead": TABLEAU_10_COLORS["purple"],
+    "SciPy-Nelder-Mead": TABLEAU_10_COLORS["brown"],
+    "Pounders": TABLEAU_10_COLORS["gray"],
+}
+
+PARALLEL_COLOR_UPDATES = {
+    "Tranquilo-LS": TABLEAU_10_COLORS["blue-75"],
+    "Tranquilo-LS (2 cores)": TABLEAU_10_COLORS["blue-55"],
+    "Tranquilo-LS (4 cores)": TABLEAU_10_COLORS["blue-35"],
+    "Tranquilo-LS (8 cores)": TABLEAU_10_COLORS["blue-15"],
+}
+
+NOISY_COLOR_UPDATES = {
+    "DFO-LS (3 evals)": TABLEAU_10_COLORS["green-25"],
+    "DFO-LS (5 evals)": TABLEAU_10_COLORS["green-50"],
+    "DFO-LS (10 evals)": TABLEAU_10_COLORS["green-75"],
+}
+
+COLORS = {
+    "scalar_benchmark": BASE_COLORS,
+    "ls_benchmark": BASE_COLORS,
+    "parallel_benchmark": {**BASE_COLORS, **PARALLEL_COLOR_UPDATES},
+    "noisy_benchmark": {**BASE_COLORS, **NOISY_COLOR_UPDATES},
+    "scalar_vs_ls_benchmark": BASE_COLORS,
+}
+
+# Line width
+# ======================================================================================
+DEFAULT_LINE_WIDTH = 1.5
+
+LINE_WIDTH_UPDATES = {
+    "parallel_benchmark": {
+        "Tranquilo-LS (2 cores)": 1.6,
+        "Tranquilo-LS (4 cores)": 1.7,
+        "Tranquilo-LS (8 cores)": 1.8,
+    },
+    "noisy_benchmark": {
+        "DFO-LS (5 evals)": 1.6,
+        "DFO-LS (10 evals)": 1.7,
+    },
+}
+
+# Legend
+# ======================================================================================
+
+LEGEND_LABEL_ORDER = {
+    "scalar_benchmark": [
+        "Tranquilo-Scalar",
+        "NAG-BOBYQA",
+        "NlOpt-BOBYQA",
+        "NlOpt-Nelder-Mead",
+        "SciPy-Nelder-Mead",
+    ],
+    "ls_benchmark": [
+        "Tranquilo-LS",
+        "DFO-LS",
+        "Pounders",
+    ],
+    "parallel_benchmark": [
+        "Tranquilo-LS",
+        "Tranquilo-LS (2 cores)",
+        "Tranquilo-LS (4 cores)",
+        "Tranquilo-LS (8 cores)",
+        "DFO-LS",
+    ],
+    "noisy_benchmark": [
+        "DFO-LS (3 evals)",
+        "DFO-LS (5 evals)",
+        "DFO-LS (10 evals)",
+        "Tranquilo-LS",
+    ],
+    "scalar_vs_ls_benchmark": [
+        "Tranquilo-LS",
+        "DFO-LS",
+        "Pounders",
+        "Tranquilo-Scalar",
+        "NlOpt-BOBYQA",
+        "NlOpt-Nelder-Mead",
+    ],
+}
+
+# Font
+# ======================================================================================
+
+matplotlib.rcParams["font.size"] = 10
+matplotlib.rcParams["font.sans-serif"] = "Fira Sans"
+matplotlib.rcParams["font.family"] = "sans-serif"
+
+
+# ======================================================================================
+# Plotting function
+# ======================================================================================
+
+
+def plot_benchmark(data, plot, benchmark):
+    """Create the base matplotlib figure.  # noqa: D406, D407, D400
+
+    Args:
+        data (dict): Dictionary containing the data to plot. Keys represent a single
+            line in the plot. The values are dictionaries with keys "x" and "y".
+        plot (str): Name of the plot to create. Must be in {"deviation_plot",
+            "profile_plot"}.
+        benchmark (str): Name of the benchmark.
+
+    Returns:
+        matplotlib.figure.Figure: The matplotlib figure.
+
+    """
+    x_range = X_RANGE[plot][benchmark]
+    legend_label_order = LEGEND_LABEL_ORDER[benchmark]
+
+    # Update label names
+    data = {LABELS[name]: line for name, line in data.items()}
+
+    # Create matplotlib base figure
+    fig, ax = plt.subplots(
+        figsize=(_cm_to_inch(FIGURE_WIDTH_IN_CM), _cm_to_inch(FIGURE_HEIGHT_IN_CM))
+    )
+    for name, line in data.items():
+        lw = LINE_WIDTH_UPDATES.get(benchmark, {}).get(name, DEFAULT_LINE_WIDTH)
+        ax.plot(
+            line["x"],
+            line["y"],
+            label=name,
+            color=COLORS[benchmark][name],
+            linewidth=lw,
+        )
+
+    # Remove top and right border (spine)
+    ax.spines[["right", "top"]].set_visible(False)
+
+    # Set color of left and bottom spine
+    ax.spines[["left", "bottom"]].set_color(DARK_GRAY)
+
+    # Update ticks
+    ax.tick_params(direction="in", width=0.75, colors=DARK_GRAY)
+
+    # Update legend
+    handles, labels = _get_sorted_legend_handles_and_labels(ax, legend_label_order)
+    ax.legend(
+        handles=_row_to_col_ordering(handles, ncol=3),
+        labels=_row_to_col_ordering(labels, ncol=3),
+        frameon=False,
+        loc="upper center",
+        ncol=3,
+        bbox_to_anchor=(0.5, -0.15),
+        labelcolor=DARK_GRAY,
+    )
+
+    # Update axes
+    ax.set_xlabel(AXIS_LABELS[plot]["xlabel"], color=DARK_GRAY)
+    ax.set_ylabel(AXIS_LABELS[plot]["ylabel"], color=DARK_GRAY)
+    ax.xaxis.label.set_color(DARK_GRAY)
+    ax.yaxis.label.set_color(DARK_GRAY)
+    ax.set_xlim(*x_range)
+    ax.set_ylim(0, 1)
+
+    return fig
+
+
+def _cm_to_inch(cm):
+    return cm * 0.393701
+
+
+def _get_sorted_legend_handles_and_labels(ax, sorted_labels=None):
+    handles, labels = ax.get_legend_handles_labels()
+    sorted_labels = labels if sorted_labels is None else sorted_labels
+    sorted_handles = [handles[labels.index(label)] for label in sorted_labels]
+    return sorted_handles, sorted_labels
+
+
+def _row_to_col_ordering(items, ncol):
+    """Reorder frm row-major to column-major ordering.
+
+    Transforms a list of items that are designed to be filled in a row-major order
+    into a list of items that are designed to be filled in a column-major order.
+
+    Taken from: https://stackoverflow.com/a/10101532.
+
+    """
+    return itertools.chain(*[items[i::ncol] for i in range(ncol)])