Skip to content

Coming from ggplot2 or seaborn

anytrellis builds a plot as ggplot2 does, a layer at a time, and names its facets as seaborn's figure-level functions do. The notebooks 03_ggplot2_style and 04_seaborn_style redo well-known figures of both.

ggplot2

ggplot2 anytrellis
facet_wrap(~ a, ncol = 3) Trellis(df, wrap="a", cols=3)
facet_grid(a ~ b) Trellis(df, row="a", col="b")
facet_grid(a ~ b, margins = TRUE) Trellis(df, row="a", col="b", margins=True)
scales = "free_y" / "free" share_y=False / share_x=False, share_y=False
geom_point(), geom_line(), geom_col() .points(), .line(), .bars()
geom_area(), geom_ribbon(), geom_tile(), geom_text() .area(), .band(), .rect(), .text()
geom_hline(), geom_vline() .rule(y=...), .rule(x=...)
geom_histogram(), geom_boxplot(), geom_violin(), geom_density() .hist(), .box(), .violin(), .density()
geom_errorbar(), geom_pointrange(), stat_summary() .errorbar()
geom_step() .line(step="post")
geom_smooth(), geom_smooth(method = "lm") .smooth(), .smooth(method="linear")
aes(colour = g), aes(group = g) color="g", group="g"
aes(colour = factor(cyl)) .as_category("cyl"), then color="cyl"
scale_y_log10() Trellis(..., y_scale="log")
position = "dodge" .bars(position="dodge")
position = "fill" .bars(position="fill"), also .area(), .hist()
geom_col() with a discrete y, coord_flip() .bars(x="value", y="category")
geom_jitter(), sns.stripplot() .points(x="category", y="value", jitter=0.4)
geom_hline(data = targets, aes(yintercept = target)) .rule(targets, y="target")
gghighlight() a layer with facet="exclude"
fct_reorder(f, x, median) .order("f", by="median", of="x")
labs(title =, subtitle =, caption =, x =, y =) Trellis(..., title=, subtitle=, caption=, x_title=, y_title=)
labs(dictionary =) (4.0) Trellis(..., labels={"mpg": "Fuel use (mpg)"})
scale_y_continuous(limits = c(0, NA)), coord_cartesian() .limits(y=(0, None))
scale_x_continuous(labels = scales::label_comma()), scale_x_date(date_labels =) .format(x=",.0f"), .format(x="%b %Y")
theme(legend.position = "right"), guides(colour = "none") .legend("right"), .legend("none")
scale_colour_viridis_c(option = "cividis"), scale_colour_distiller(palette = "Blues") .colormap("cividis"), .colormap("Blues")
scale_colour_gradient2(midpoint = 0) .colormap(scale="diverging", midpoint=0)
scale_colour_binned(), scale_colour_viridis_b() .colormap(scale="binned")
scale_colour_manual(values = c(USA = "red")) Trellis(..., palette={"USA": "red"})
ggplot(penguins, aes(flipper_length_mm, body_mass_g, colour = sex)) +
  geom_point() + facet_wrap(~ species)
from anytrellis import Trellis
from anytrellis.datasets import load

penguins = load("penguins")
plot = Trellis(penguins, wrap="species", panel_height=170)
plot.points(x="flipper_length_mm", y="body_mass_g", color="sex")
plot

What differs:

  • Columns by name, not aes(): each mark takes x=, y= and color= as column names.
  • No stats on plain marks. bars() draws the heights you give it; hist() counts. Summarise with your table library first, as for geom_col().
  • Two looks, no theme objects. theme="minimal" (the default, close to theme_minimal()) or theme="classic" (close to theme_grey()); colours come from palette= and color=; the plot follows the notebook's light or dark theme.
  • Interactive: every plot zooms, pans, brushes and hands its selection to Python.

seaborn

seaborn's relplot, displot, catplot, lmplot and FacetGrid take col=, row=, col_wrap= and hue=:

seaborn anytrellis
col="a", row="b" col="a", row="b"
col="a", col_wrap=3 wrap="a", cols=3
hue="g" color="g"
facet_kws={"sharey": False} share_y=False
relplot(kind="scatter"), relplot(kind="line") .points(), .line()
displot(kind="hist"), displot(kind="kde") .hist(), .density()
catplot(kind="box"), catplot(kind="bar") .box(), .bars() (of means you compute)
catplot(kind="violin") .violin()
catplot(kind="point"), pointplot(errorbar="ci") .errorbar(), .errorbar(extent="ci")
lineplot(drawstyle="steps-post") .line(step="post")
pairplot(df, hue="g"), pairplot(corner=True) pairs(df, color="g"), pairs(df, corner=True)
lmplot .points() and .smooth(method="linear")
bw_adjust= bw_adjust=
sns.lmplot(data=penguins, x="flipper_length_mm", y="body_mass_g", hue="sex", col="species")
known = penguins.drop_nulls("sex")
plot = (
    Trellis(known, col="species", panel_height=220)
    .points(x="flipper_length_mm", y="body_mass_g", color="sex", size=5)
    .smooth(x="flipper_length_mm", y="body_mass_g", color="sex", method="linear")
)
plot

catplot(kind="bar") computes means and confidence intervals itself; with anytrellis, compute the means with Polars or pandas and draw them with bars().