Styles, colour choices, and saving figures

Understanding Styles, colour choices, and saving figures

Every default in Matplotlib comes from one dictionary, plt.rcParams: line widths, font sizes, whether grids are drawn, the colour cycle, even the default figure size. A style such as "ggplot" or "grayscale" is nothing more magical than a named batch of edits to that dictionary, so plt.style.use('ggplot') changes future figures but never touches ones already drawn. Precedence runs in one direction: an explicit keyword like color="#0072B2" on a plot call beats the style, the style beats the built-in defaults. Because plt.style.use is global and sticky, prefer plt.style.context(...) as a with-block when you only want the look changed for one figure.

Colour is doing one of three different jobs, and choosing badly is a correctness problem, not a taste problem. For unordered categories you want colours that are merely distinguishable, which is what the property cycle gives you; C0 through C9 are not fixed colours but positions in the current cycle, so they change when the style changes, while tab:blue and #1f77b4 are fixed. For a quantity that has an order you want a sequential colormap like viridis, which is perceptually uniform, so equal steps in the data look like equal steps in brightness; rainbow maps like jet invent bright bands that read as features in the data that are not there. For data with a meaningful centre, such as anomalies around zero, use a diverging map like coolwarm and centre the norm on that value.

Saving is a fresh render, not a screenshot: savefig redraws the figure into the requested backend, so the file can differ from what a window showed. For raster formats the pixel size is figsize multiplied by dpi, which is why figsize=(4, 3) at dpi=200 gives 800x600; for PDF and SVG the geometry is vector and dpi only affects embedded images. bbox_inches="tight" recomputes the bounding box from the artists actually drawn, which is how you stop long tick labels and legends from being clipped, at the price of no longer knowing the exact output size in advance. Call savefig before show, because some interactive backends hand the figure to the GUI and clear it, leaving you with a blank file.

import matplotlib
matplotlib.use("Agg")            # write files, no GUI window needed
import matplotlib.pyplot as plt
import os

x = [0, 1, 2, 3, 4]

with plt.style.context("ggplot"):
    fig, ax = plt.subplots(figsize=(4, 3))
    ax.plot(x, [v * v for v in x], color="#0072B2", linewidth=2, label="v squared")
    ax.plot(x, [v * 3 for v in x], color="#D55E00", linestyle="--", label="3v")
    ax.set_title("explicit colours override the style cycle")
    ax.legend()
    fig.savefig("plot.png", dpi=150, bbox_inches="tight")
    fig.savefig("plot.svg")
    plt.close(fig)

print("png written:", os.path.exists("plot.png"))
with open("plot.svg") as f:
    print("svg begins with:", f.read(5))
print("cycle outside the context:", plt.rcParams["axes.prop_cycle"].by_key()["color"][:3])
print("grid outside the context:", plt.rcParams["axes.grid"])

Worked examples

import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt

print("start grid:", plt.rcParams["axes.grid"])

with plt.style.context("ggplot"):
    print("inside ggplot:", plt.rcParams["axes.grid"])

print("after context:", plt.rcParams["axes.grid"])

plt.rcParams["axes.grid"] = True
plt.rcParams["figure.dpi"] = 110
print("after manual set:", plt.rcParams["axes.grid"], plt.rcParams["figure.dpi"])

plt.rcdefaults()
print("after rcdefaults:", plt.rcParams["axes.grid"], plt.rcParams["figure.dpi"])

import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
from matplotlib import colors

viridis = plt.get_cmap("viridis")
print("viridis low: ", colors.to_hex(viridis(0.0)))
print("viridis high:", colors.to_hex(viridis(1.0)))
print("C0 ->", colors.to_hex("C0"))
print("tab:cyan ->", colors.to_hex("tab:cyan"))
print("grey '0.5' ->", colors.to_hex("0.5"))

import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import struct

def png_size(path):
    with open(path, "rb") as f:
        header = f.read(24)
    return struct.unpack(">II", header[16:24])

fig, ax = plt.subplots(figsize=(4, 2))
ax.plot([1, 3, 2])
fig.savefig("a.png", dpi=100)
fig.savefig("b.png", dpi=200)
plt.close(fig)

print("dpi=100:", png_size("a.png"))
print("dpi=200:", png_size("b.png"))
print("figsize inches:", (4.0, 2.0))

import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import struct

def png_size(path):
    with open(path, "rb") as f:
        return struct.unpack(">II", f.read(24)[16:24])

fig, ax = plt.subplots(figsize=(4, 3))
ax.plot([0, 1], [0, 1], color="tab:green")
ax.set_ylabel("a deliberately long axis label")

fig.savefig("plain.png", dpi=100)
fig.savefig("tight.png", dpi=100, bbox_inches="tight", transparent=True)
plt.close(fig)

print("plain:", png_size("plain.png"))
print("tight is smaller than plain:", png_size("tight.png") < png_size("plain.png"))

Important notes

Common mistakes

Try it yourself