TopolearnA consistent theme and the right palette type make charts readable and honest with almost no code.
sns.set_theme() applies seaborn's default look to all subsequent Matplotlib figures: the "darkgrid" style, the "notebook" context and the "deep" palette. Its arguments change each part:
style=:"darkgrid","whitegrid","dark","white","ticks".sns.despine()removes the top and right spines.context=:"paper","notebook","talk","poster"scale fonts and line widths for the destination;font_scale=fine-tunes.palette=: the default color cycle.
sns.axes_style() and sns.plotting_context() also work as context managers for temporary changes.
Pick a palette by data type:
- Qualitative for categories:
"deep","colorblind","Set2","tab10". - Sequential for ordered numbers:
"rocket","mako","viridis","Blues"; seaborn uses these automatically for numerichue. - Diverging for values around a center:
"vlag","icefire","RdBu".
sns.color_palette("mako", 5) returns a list of RGB colors you can reuse anywhere, and as_cmap=True returns a Matplotlib colormap. You can also pass a dict, palette={"control": "gray", "treatment": "C1"}, to pin specific colors to specific categories so they stay consistent across charts.
Use the "colorblind" palette in anything others will read, and avoid more than about eight qualitative colors; beyond that, colors become hard to tell apart.
Example
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import seaborn as sns
from matplotlib.colors import to_hex
sns.set_theme(style="whitegrid", context="talk", palette="colorblind")
print(plt.rcParams["font.size"]) # 18.0 (talk context is larger)
pal = sns.color_palette("mako", 3)
print(len(pal), to_hex(pal[0])) # 3 #3e356b
cmap = sns.color_palette("vlag", as_cmap=True)
print(type(cmap).__name__) # ListedColormap
fig, ax = plt.subplots()
sns.barplot(x=["control", "treatment"], y=[3.1, 3.8],
hue=["control", "treatment"],
palette={"control": "gray", "treatment": "C1"}, ax=ax)
sns.despine()
print(ax.spines["top"].get_visible()) # False
fig.savefig("themed.png")Remember
set_themesets style, context and palette together- Contexts scale everything for paper, notebook, talk or poster
- Qualitative palettes for categories, sequential for magnitudes, diverging around a center
Watch out for
- Using a sequential palette for unordered categories, implying a ranking that does not exist.