Scientific Figure Codebases.

scientific_figure.py

import torch
from typing import List, Tuple
import matplotlib.pyplot as plt
from matplotlib import ticker
from matplotlib import font_manager

# more details about matplotlib usage at
# https://wizardforcel.gitbooks.io/matplotlib-user-guide/content/

def color(*items):
    return [item / 255 for item in items]

# All colors in the followings are from
# https://sci-figure-colors.unvs.cc/%E7%A7%91%E7%A0%94%E8%AE%BA%E6%96%87%E6%8F%92%E5%9B%BE%E9%85%8D%E8%89%B2.pdf


COLORS = {}

# No.002
COLORS[2] = {
    "red": color(223, 122, 94),
    "blue": color(60, 64, 91),
    "green": color(130, 178, 194),
    "yellow": color(240, 210, 134)
}

# No.006
COLORS[6] = {
    "red": color(230, 111, 81),
    "blue": color(38, 70, 83),
    "light_green": color(138, 176, 125),
    "green": color(41, 157, 143),
    "dark_green": color(40, 114, 113),
    "yellow": color(232, 197, 107),
    "orange": color(243, 162, 97)
}

# No.007
COLORS[7] = {
    "dark_red": color(168, 3, 38),
    "red": color(236, 93, 59),
    "orange": color(253, 185, 107),
    "dark_blue": color(57, 81, 162),
    "blue": color(114, 170, 207),
    "light_blue": color(202, 232, 242),
}


def show_available_font():
    # more details at https://cloud.tencent.com/developer/article/1761532
    print("All available font in matplotlib are as follows:")
    for font in font_manager.fontManager.ttflist:
        print(font.name, '-', font.fname)


def set_global_matplotlib_font(fontfamily: str):
    plt.rcParams['font.sans-serif'] = fontfamily


def add_temp_font_in_matplotlib(fontpath):
    font_manager.fontManager.addfont(fontpath)


def to_percent(value, position):
    return f"{value*100:.0f}%"


def plot_line(
    Xs: List[List[float]],
    Ys: List[List[float]],
    labels: List[str],
    x_labels: List[str],
    y_labels: List[str],
    colors: List[Tuple[float]],
    title: str,
    subtitles: List[str],
    n_rows: int = 1,
    xlims: List[Tuple[float]] = None,
    ylims: List[Tuple[float]] = None,
    figsize: List[int] = None,
    grid: bool = False,
    top_right_visible: bool = True,
    linewidth: int = 3,
    fontsize: int = 20,
    yaxis_formatters=None,
    save_path: str = "figure.png",
):
    assert len(Xs) == len(Ys)
    assert len(Xs) == len(colors)
    assert len(Xs[0]) == len(colors[0])
    if labels is not None:
        assert len(Xs[0]) == len(labels)
    else:
        labels = [None] * len(Xs[0])

    if xlims is None:
        xlims = [None] * len(Xs)
    if ylims is None:
        ylims = [None] * len(Xs)
    if yaxis_formatters is None:
        yaxis_formatters = [None] * len(Xs)

    assert len(Xs) % n_rows == 0

    n_columns = len(Xs) // n_rows
    if figsize is None:
        figsize = (n_columns*9, n_rows*5)

    fig = plt.figure(figsize=figsize)

    for i, (_X, _Y, x_label, y_label, xlim, ylim, subtitle, yaxis_formatter, subcolors) in \
            enumerate(zip(Xs, Ys, x_labels, y_labels, xlims, ylims, subtitles, yaxis_formatters, colors), start=1):
        ax = fig.add_subplot(n_rows, n_columns, i)
        for j, (x, y,  label, color) in \
                enumerate(zip(_X, _Y,  labels, subcolors)):
            ax.plot(x, y, color=color, linestyle='-', linewidth=linewidth, label=label)

        ax.tick_params(axis='both', which='major', labelsize=fontsize)
        ax.tick_params(axis='both', which='minor', labelsize=fontsize)
        ax.set_xlabel(x_label, fontsize=fontsize)
        ax.set_ylabel(y_label, fontsize=fontsize)
        ax.set_title(subtitle, fontsize=fontsize)
        ax.set_xlim(xlim)
        ax.set_ylim(ylim)
        ax.legend()

        if yaxis_formatter:
            ax.yaxis.set_major_formatter(ticker.FuncFormatter(yaxis_formatter))

        if grid:
            ax.grid(color='gray', linestyle='--', linewidth=0.5)

        if not top_right_visible:
            ax.spines['top'].set_visible(False)
            ax.spines['right'].set_visible(False)

        ax.legend(fontsize=fontsize)

    fig.suptitle(title, fontsize=fontsize+10)
    fig.tight_layout()

    if save_path.endswith("pdf"):
        fig.savefig(save_path, format="pdf", dpi=600)
    elif save_path.endswith("svg"):
        fig.savefig(save_path, format="svg", dpi=600)
    else:
        fig.savefig(save_path)


baseline = torch.load("baseline_metrics.pt")
oursv1 = torch.load("our_metrics.pt")
oursv2 = torch.load("oursv2_metrics.pt")

x = list(range(1, 91))

show_available_font()
set_global_matplotlib_font("Times New Roman")

plot_line(
    Xs=[[x]*3]*4,
    Ys=[
        [baseline["train/loss"], oursv1["train/loss"], oursv2["train/loss"]],
        [baseline["eval/loss"], oursv1["eval/loss"], oursv2["eval/loss"]],
        [baseline["train/top1_acc"], oursv1["train/top1_acc"], oursv2["train/top1_acc"]],
        [baseline["eval/top1_acc"], oursv1["eval/top1_acc"], oursv2["eval/top1_acc"]]
    ],
    labels=["baseline", "Ours-A", "Ours-B"],
    x_labels=["Epoch"]*4,
    y_labels=["Loss", "Loss", 'Acc. (%)', 'Acc. (%)'],
    xlims=[[0, 90]]*4,
    ylims=[[0, 6], [0, 6], [0, 1], [0, 1]],
    colors=[[COLORS[6]["dark_green"], COLORS[7]["dark_blue"], COLORS[7]["dark_red"]]]*4,
    title=f"Evolution of Loss/Accuracy",
    subtitles=["Training Loss", "Eval Loss", "Training Top-1 Acc.", "Eval Top-1 Acc."],
    n_rows=2,
    grid=True,
    top_right_visible=False,
    yaxis_formatters=[None, None, to_percent, to_percent]
)