A Full Information to Seaborn

Picture by Editor

# Introduction

 
Seaborn is a statistical visualization library for Python that sits on high of Matplotlib. It offers you clear defaults, tight integration with Pandas DataFrames, and high-level features that scale back boilerplate. When you already know Matplotlib and wish quicker, extra informative plots, this information is for you.

The main target right here is intermediate to superior utilization. You’ll work with relational, categorical, distribution, and regression plots, then transfer into grid layouts and matrix visuals that reply actual analytical questions. Count on quick code blocks, exact explanations, and sensible parameter selections that have an effect on readability and accuracy.

What this information covers:

• Arrange themes and palettes you’ll be able to reuse throughout initiatives
• Plots that matter for evaluation: scatterplot, lineplot, boxplot, violinplot, histplot, kdeplot, regplot, lmplot
• Excessive-dimensional layouts with FacetGrid, PairGrid, relplot, and catplot
• Correlation and heatmaps with right shade scales, masking, and annotation
• Exact management by way of Matplotlib hooks for titles, ticks, legends, and annotations
• Efficiency suggestions for big datasets and fixes for frequent pitfalls

You’ll be taught when to make use of confidence intervals, learn how to handle legends in crowded visuals, learn how to preserve class colours constant, and when to modify again to Matplotlib for high quality management. The objective is obvious, correct plots that talk findings with out further work.

# Setup and Styling Baseline

 
This part units a constant visible baseline so each plot within the article appears to be like skilled and export-ready. We are going to set up, import, set a worldwide theme, select sensible palettes, and lock in determine sizing and DPI for clear outputs.

// Set up and import

Use a clear atmosphere and set up Seaborn and Matplotlib.

pip set up seaborn matplotlib

Customary imports:

import seaborn as sns
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd

Two fast checks that assist keep away from surprises:

// Mission-wide theme in a single line

Set a default model as soon as, then give attention to the evaluation as a substitute of fixed styling tweaks.

sns.set_theme(
    context="speak",      # textual content measurement scaling: paper, pocket book, speak, poster
    model="whitegrid",   # clear background with gentle grid
    palette="deep"       # readable, colorblind conscious categorical palette
)

Why this issues:

• context="speak" offers readable axis labels and titles for slides and experiences
• model="whitegrid" improves worth studying for line and bar plots with out heavy visible noise
• palette="deep" offers distinct class colours that maintain up when printed or projected

You’ll be able to override any of those per plot, however setting them globally retains the look uniform.

// Palettes you’ll truly use

Select palettes that talk the information kind. Use discrete palettes for classes and colormaps for steady values.

1. Viridis for steady scales

# Discrete colours for classes
cats = sns.color_palette("viridis", n_colors=5)

# Steady colormap for heatmaps and densities
viridis_cmap = sns.color_palette("viridis", as_cmap=True)

• Viridis preserves element throughout gentle and darkish backgrounds and is perceptually uniform
• Use n_colors= for discrete classes. Use as_cmap=True when mapping a numeric vary

2. Cubehelix for ordered classes or low-ink plots

# Mild-to-dark sequence that prints properly
dice = sns.cubehelix_palette(
    begin=0.5,    # hue begin
    rot=-0.75,    # hue rotation
    gamma=1.0,    # depth curve
    gentle=0.95,
    darkish=0.15,
    n_colors=6
)

Cubehelix stays readable in grayscale and helps ordered classes the place development issues.

3. Mix a customized model ramp

# Mix two model colours right into a easy ramp
mix = sns.blend_palette(["#0F766E", "#60A5FA"], n_colors=7, as_cmap=False)

# When you want a steady colormap as a substitute
blend_cmap = sns.blend_palette(["#0F766E", "#60A5FA"], as_cmap=True)

Mixing helps align visuals with a design system whereas protecting numerical gradients constant.

Set a palette globally if you decide to a scheme for a complete determine or report

sns.set_palette(cats)     # or dice, or mix

Preview a palette shortly

sns.palplot(cats)
plt.present()

// Determine sizing and DPI for export

Management measurement and backbone from the begin to keep away from fuzzy labels or cramped axes.
Set a wise default as soon as

# International defaults through Matplotlib rcParams
plt.rcParams["figure.figsize"] = (8, 5)    # width, peak in inches
plt.rcParams["figure.dpi"] = 150           # on-screen readability with out big recordsdata

You’ll be able to nonetheless measurement particular person figures explicitly when wanted:

fig, ax = plt.subplots(figsize=(8, 5))

Save high-quality outputs

# Raster export for internet or slide decks
plt.savefig("determine.png", dpi=300, bbox_inches="tight")

# Vector export for print or journals
plt.savefig("determine.svg", bbox_inches="tight")
plt.savefig("determine.pdf", bbox_inches="tight")

• dpi=300 is an efficient goal for crisp internet photographs and displays
• bbox_inches="tight" trims empty margins, which retains multi-panel layouts compact
• Want SVG or PDF when editors will resize figures or if you want sharp textual content at any scale

# Plots That Matter for Actual Work

 
On this part, we are going to give attention to plot varieties that reply evaluation questions shortly. Every subsection explains when to make use of the plot, the important thing parameters that management that means, and a brief code pattern you’ll be able to adapt. The examples assume you already set the theme and baseline from the earlier part.

// Relational plots: scatterplot, relplot(form="line")

Use relational plots to point out relationships between numeric variables and to match teams with shade, marker, and measurement encodings. Add readability by mapping a categorical variable to hue and model, and a numeric variable to measurement.

import seaborn as sns
import matplotlib.pyplot as plt

penguins = sns.load_dataset("penguins").dropna(
    subset=["bill_length_mm", "bill_depth_mm", "body_mass_g", "species", "sex"]
)

# Scatter with a number of encodings
ax = sns.scatterplot(
    information=penguins,
    x="bill_length_mm",
    y="bill_depth_mm",
    hue="species",
    model="intercourse",
    measurement="body_mass_g",
    sizes=(30, 160),
    alpha=0.8,
    edgecolor="w",
    linewidth=0.5
)
ax.set_title("Invoice size vs depth with species, intercourse, and mass encodings")
ax.legend(title="Species")
plt.tight_layout()
plt.present()

For traces, favor the figure-level API if you want markers per group and simple faceting.

flights = sns.load_dataset("flights")  # 12 months, month, passengers

g = sns.relplot(
    information=flights,
    form="line",
    x="12 months",
    y="passengers",
    hue="month",
    markers=True,      # marker on every level
    dashes=False,      # strong traces for all teams
    peak=4, side=1.6
)
g.set_axis_labels("Yr", "Passengers")
g.determine.suptitle("Month-to-month passenger pattern by 12 months", y=1.02)
plt.present()

Notes:

• Use model for a second categorical channel when hue shouldn’t be sufficient
• Maintain alpha barely under 1.0 on dense scatters to disclose overlap
• Use sizes=(min, max) to constrain level sizes so the legend stays readable

// Categorical plots: boxplot, violinplot, barplot

Categorical plots present distributions and group variations. Select field or violin if you care about unfold and outliers. Select bar if you need aggregated values with intervals.

import numpy as np
suggestions = sns.load_dataset("suggestions")

# Boxplot: strong abstract of unfold
ax = sns.boxplot(
    information=suggestions,
    x="day",
    y="total_bill",
    hue="intercourse",
    order=["Thur", "Fri", "Sat", "Sun"],
    dodge=True,
    showfliers=False
)
ax.set_title("Complete invoice by day and intercourse (boxplot, fliers hidden)")
plt.tight_layout()
plt.present()

# Violin: form of the distribution with quartiles
ax = sns.violinplot(
    information=suggestions,
    x="day",
    y="total_bill",
    hue="intercourse",
    order=["Thur", "Fri", "Sat", "Sun"],
    dodge=True,
    internal="quartile",
    minimize=0,
    scale="width"
)
ax.set_title("Complete invoice by day and intercourse (violin with quartiles)")
plt.tight_layout()
plt.present()

# Bar: imply tip with percentile intervals
ax = sns.barplot(
    information=suggestions,
    x="day",
    y="tip",
    hue="intercourse",
    order=["Thur", "Fri", "Sat", "Sun"],
    estimator=np.imply,
    errorbar=("pi", 95),   # percentile interval for skewed information
    dodge=True
)
ax.set_title("Imply tip by day and intercourse with 95% PI")
plt.tight_layout()
plt.present()

Notes:

• order fixes class sorting for constant comparisons
• For big samples the place intervals add noise, set errorbar=None (or ci=None on older Seaborn)
• Conceal fliers on boxplots when excessive factors distract from the group comparability

// Distribution plots:histplot

Distribution plots reveal form, multimodality, and group variations. Use stacking if you need totals, and fill if you need composition.

# Single distribution with a easy density overlay
ax = sns.histplot(
    information=penguins,
    x="body_mass_g",
    bins=30,
    kde=True,
    aspect="step"
)
ax.set_title("Physique mass distribution with KDE")
plt.tight_layout()
plt.present()

# Grouped comparability: composition throughout species
ax = sns.histplot(
    information=penguins,
    x="body_mass_g",
    hue="species",
    bins=25,
    a number of="fill",    # fraction per bin (composition)
    aspect="step",
    stat="proportion",
    common_norm=False
)
ax.set_title("Physique mass composition by species")
plt.tight_layout()
plt.present()

# Grouped comparability: whole counts by stacking
ax = sns.histplot(
    information=penguins,
    x="body_mass_g",
    hue="species",
    bins=25,
    a number of="stack",
    aspect="step",
    stat="rely"
)
ax.set_title("Physique mass counts by species (stacked)")
plt.tight_layout()
plt.present()

Notes:

• Use a number of="fill" to match relative composition throughout bins
• Use common_norm=False when teams differ in measurement and also you need within-group densities
• Select aspect="step" for clear edges and simple overlaying

// Regression plots: regplot, lmplot

Regression plots add fitted relationships and intervals. Use regplot for a single axes. Use lmplot if you want hue, row, or col faceting with out handbook grid work.

Let’s check out an lmplot. Simply make sure the dataset has no lacking values within the mapped columns.

penguins = sns.load_dataset("penguins").dropna(
    subset=["bill_length_mm", "bill_depth_mm", "species", "sex"]
)

g = sns.lmplot(
    information=penguins,
    x="bill_length_mm",
    y="bill_depth_mm",
    hue="species",
    col="intercourse",
    peak=4,
    side=1,
    scatter_kws=dict(s=25, alpha=0.7),
    line_kws=dict(linewidth=2)
)
g.set_titles(col_template="{col_name}")
g.determine.suptitle("Invoice dimensions by species and intercourse", y=1.02)
plt.present()

Notes:

• On newer Seaborn variations, favor errorbar=("ci", 95) on features that assist it. If ci remains to be accepted in your model, you’ll be able to preserve utilizing it for now.
• When you see related errors, verify for different unique pairs like lowess=True, logistic=True, or logx=True used collectively.

// Interval selections on massive information

On huge samples, interval bands can obscure the sign. Two choices enhance readability:

• Use percentile intervals for skewed distributions:

sns.barplot(information=suggestions, x="day", y="tip", errorbar=("pi", 95))

 

• Take away intervals solely when variation is already apparent:

sns.lineplot(information=flights, x="12 months", y="passengers", errorbar=None)
# or on older variations:
sns.lineplot(information=flights, x="12 months", y="passengers", ci=None)

 

Guideline:

• Want errorbar=("pi", 95) for skewed or heavy-tailed information
• Want errorbar=None (or ci=None) when the viewers cares extra about pattern form than exact uncertainty on a really massive N

# Excessive-Dimensional Views with Grids

 
Grids provide help to examine patterns throughout teams with out handbook subplot juggling. You outline rows, columns, and shade as soon as, then apply a plotting perform to every subset. This retains construction constant and makes variations apparent.

// FacetGrid and catplot/ relplot

Use a FacetGrid if you need full management over what will get mapped to every aspect. Use catplot and relplot if you desire a fast, figure-level API that builds the grid for you. The core thought is identical: break up information by row, col, and shade with hue.
Earlier than the code: preserve aspect counts practical. 4 to 6 small multiples are simple to scan. Past that, wrap columns or filter classes. Management sharing with sharex and sharey so comparisons stay legitimate.

import seaborn as sns
import matplotlib.pyplot as plt

suggestions = sns.load_dataset("suggestions").dropna()

# 1) Full management with FacetGrid + regplot
g = sns.FacetGrid(
    information=suggestions,
    row="time",                # Lunch vs Dinner
    col="day",                 # Thur, Fri, Sat, Solar
    hue="intercourse",                 # Male vs Feminine
    margin_titles=True,
    sharex=True,
    sharey=True,
    peak=3,
    side=1
)
g.map_dataframe(
    sns.regplot,
    x="total_bill",
    y="tip",
    scatter_kws=dict(s=18, alpha=0.6),
    line_kws=dict(linewidth=2),
    ci=None
)
g.add_legend(title="Intercourse")
g.set_axis_labels("Complete invoice", "Tip")
g.fig.suptitle("Tipping patterns by day and time", y=1.02)
plt.present()

# 2) Fast grids with catplot (constructed on FacetGrid)
sns.catplot(
    information=suggestions,
    form="field",
    x="day", y="total_bill",
    row="time", hue="intercourse",
    order=["Thur", "Fri", "Sat", "Sun"],
    peak=3, side=1.1, dodge=True
).set_axis_labels("Day", "Complete invoice")
plt.present()

# 3) Fast relational grids with relplot
penguins = sns.load_dataset("penguins").dropna()
sns.relplot(
    information=penguins,
    form="scatter",
    x="bill_length_mm", y="bill_depth_mm",
    row="intercourse", col="island", hue="species",
    peak=3.2, side=1
)
plt.present()

Key factors:

• Use order to repair class sorting
• Use col_wrap when you might have one aspect dimension with many ranges
• Add a suptitle to summarize the comparability; preserve axis labels constant throughout aspects

// PairGrid and pairplot

Pairwise plots reveal relationships throughout many numeric variables. pairplot is the quick path. PairGrid offers you per-region management. For dense datasets, restrict variables and take into account nook=True to drop redundant higher panels.

Earlier than the code: select variables which can be informative collectively. Combine scales solely when you might have a motive, then standardize or log-transform first.

# Fast pairwise view
num_cols = ["bill_length_mm", "bill_depth_mm", "flipper_length_mm", "body_mass_g"]
sns.pairplot(
    information=penguins[num_cols + ["species"]].dropna(),
    vars=num_cols,
    hue="species",
    nook=True,           # solely decrease triangle + diagonal
    diag_kind="hist",      # or "kde"
    plot_kws=dict(s=18, alpha=0.6),
    diag_kws=dict(bins=20, aspect="step")
)
plt.present()

Ideas:

• nook=True reduces litter and quickens rendering
• Maintain marker measurement modest so overlaps stay readable
• For very completely different scales, apply np.log10 to skewed measures earlier than plotting

// Combined layers on a PairGrid

A blended mapping helps you examine scatter patterns and density construction in a single view. Use scatter on the higher triangle, bivariate KDE on the decrease triangle, and histograms on the diagonal. This mixture is compact and informative.
Earlier than the code: density layers can get heavy. Cut back ranges and keep away from too many bins. Add a legend as soon as and preserve it outdoors the grid if area is tight.

from seaborn import PairGrid

g = PairGrid(
    information=penguins[num_cols + ["species"]].dropna(),
    vars=num_cols,
    hue="species",
    peak=2.6, side=1
)

# Higher triangle: scatter
g.map_upper(
    sns.scatterplot,
    s=16, alpha=0.65, linewidth=0.3, edgecolor="w"
)

# Decrease triangle: bivariate KDE
g.map_lower(
    sns.kdeplot,
    fill=True, thresh=0.05, ranges=5
)

# Diagonal: histograms
g.map_diag(
    sns.histplot,
    bins=18, aspect="step"
)

g.add_legend(title="Species")
for ax in g.axes.flat:
    if ax shouldn't be None:
        ax.tick_params(axis="x", labelrotation=30)

g.fig.suptitle("Pairwise construction of penguin measurements", y=1.02)
plt.present()

Tips:

• Begin with 4 numeric variables. Add extra provided that every provides a definite sign
• For uneven group sizes, give attention to proportions fairly than uncooked counts if you examine distributions
• If rendering slows down, pattern rows earlier than plotting or drop fill from the KDE layer

# Correlation, Heatmaps, and Matrices

 
Correlation heatmaps are a compact approach to scan relationships throughout many numeric variables. The objective is a readable matrix that highlights actual sign, retains noise out of the way in which, and exports cleanly.

// Construct a correlation matrix and masks redundant cells

Begin by choosing numeric columns and selecting a correlation methodology. Pearson is commonplace for linear relationships. Spearman is best for ranked or monotonic patterns. A triangular masks removes duplication so the attention focuses on distinctive pairs.

import seaborn as sns
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd

# Information
penguins = sns.load_dataset("penguins").dropna()

# Select numeric columns and compute correlation
num_cols = penguins.select_dtypes(embrace="quantity").columns
corr = penguins[num_cols].corr(methodology="pearson")

# Masks the higher triangle (preserve decrease + diagonal)
masks = np.triu(np.ones_like(corr, dtype=bool))

# Heatmap with diverging palette centered at zero
ax = sns.heatmap(
    corr,
    masks=masks,
    annot=True,
    fmt=".2f",
    cmap="vlag",
    middle=0,
    vmin=-1, vmax=1,
    sq.=True,
    cbar_kws={"shrink": 0.8, "label": "Pearson r"},
    linewidths=0.5, linecolor="white"
)

ax.set_title("Correlation matrix of penguin measurements")
plt.tight_layout()
plt.present()

Notes:

• Use methodology="spearman" when variables aren’t on comparable scales or comprise outliers that have an effect on Pearson
• Maintain vmin and vmax symmetric so the colour scale treats unfavourable and constructive values equally

// Management visibility with scale and colorbar choices

As soon as the matrix is in place, tune what the reader sees. Symmetric limits, a centered palette, and a labeled colorbar stop misreads. You can too conceal weak correlations or the diagonal to cut back litter.

# Elective: conceal weak correlations under a threshold
threshold = 0.2
weak = corr.abs() < threshold
mask2 = np.triu(np.ones_like(corr, dtype=bool)) | weak  # mix masks

ax = sns.heatmap(
    corr,
    masks=mask2,
    annot=False,                 # let sturdy colours carry the message
    cmap="vlag",
    middle=0,
    vmin=-1, vmax=1,
    sq.=True,
    cbar_kws={"shrink": 0.8, "ticks": [-1, -0.5, 0, 0.5, 1], "label": "Correlation"},
    linewidths=0.4, linecolor="#F5F5F5"
)

ax.set_title("Sturdy correlations solely (|r| ≥ 0.20)")
plt.tight_layout()
plt.present()

Ideas:

• cbar_kws controls readability of the legend. Set ticks that match your viewers
• Flip annot=True again on if you want precise values for a report. Maintain it off for dashboards the place form and shade are sufficient

// Massive matrices: preserve labels and edges readable

Large matrices want self-discipline. Skinny or rotate tick labels, add grid traces between cells, and take into account reordering variables to group associated blocks. If the matrix could be very large, present each nth tick to keep away from label collisions.

# Artificial large instance: 20 numeric columns
rng = np.random.default_rng(0)
large = pd.DataFrame(rng.regular(measurement=(600, 20)),
                    columns=[f"f{i:02d}" for i in range(1, 21)])

corr_wide = large.corr()

fig, ax = plt.subplots(figsize=(10, 8), dpi=150)

hm = sns.heatmap(
    corr_wide,
    cmap="vlag",
    middle=0,
    vmin=-1, vmax=1,
    sq.=True,
    cbar_kws={"shrink": 0.7, "label": "Correlation"},
    linewidths=0.3, linecolor="white"
)

# Rotate x labels and skinny ticks
ax.set_xticklabels(ax.get_xticklabels(), rotation=40, ha="proper")
ax.set_yticklabels(ax.get_yticklabels(), rotation=0)
ax.tick_params(axis="each", labelsize=8)

# Present each 2nd tick on each axes
xt = ax.get_xticks()
yt = ax.get_yticks()
ax.set_xticks(xt[::2])
ax.set_yticks(yt[::2])

ax.set_title("Correlation matrix with tick thinning and grid traces")
plt.tight_layout()
plt.present()

When construction issues greater than precise order, strive a clustered view that teams related variables:

# Clustered matrix for sample discovery
g = sns.clustermap(
    corr_wide,
    cmap="vlag",
    middle=0,
    vmin=-1, vmax=1,
    linewidths=0.2, linecolor="white",
    figsize=(10, 10),
    cbar_kws={"shrink": 0.6, "label": "Correlation"},
    methodology="common",  # linkage
    metric="euclidean" # distance on correlations
)
g.fig.suptitle("Clustered correlation matrix", y=1.02)
plt.present()

Tips:

• Enhance determine measurement fairly than shrinking font till it turns into unreadable
• Add linewidths and linecolor to outline cell boundaries on dense matrices
• Use clustering if you wish to floor block construction. Maintain a plain ordered matrix if you want steady positions throughout experiences

# Precision Management with Matplotlib Hooks

 
Seaborn handles the heavy lifting, however ultimate polish comes from Matplotlib. These hooks allow you to set clear titles, management axes exactly, handle legends in tight areas, and annotate vital factors with out litter.

// Titles, labels, legends

Good plots learn themselves. Set titles that state the query, label axes with items, and preserve the legend compact and informative. Place the legend the place it helps the attention, not the place it hides information.

Earlier than the code: favor axis-level strategies over plt.* so settings keep hooked up to the correct subplot. Use a legend title and take into account shifting the legend outdoors the axes when you might have many teams.

import seaborn as sns
import matplotlib.pyplot as plt

penguins = sns.load_dataset("penguins").dropna()

ax = sns.scatterplot(
    information=penguins,
    x="bill_length_mm",
    y="bill_depth_mm",
    hue="species",
    model="intercourse",
    s=60,
    alpha=0.8,
    edgecolor="w",
    linewidth=0.5
)

# Titles and labels
ax.set_title("Invoice size vs depth by species")
ax.set_xlabel("Invoice size (mm)")
ax.set_ylabel("Invoice depth (mm)")

# Legend with title, positioned outdoors to the correct
leg = ax.legend(
    title="Species",
    loc="middle left",
    bbox_to_anchor=(1.02, 0.5),   # outdoors the axes
    frameon=True,
    borderaxespad=0.5
)

# Elective legend styling
for textual content in leg.get_texts():
    textual content.set_fontsize(9)
leg.get_title().set_fontsize(10)

plt.tight_layout()
plt.present()

Notes

• bbox_to_anchor offers you high quality management over legend placement outdoors the axes
• Maintain legend fonts barely smaller than axis tick labels to cut back visible weight
• When you want a customized legend order, move hue_order= within the plotting name

// Axis management

Axis limits, ticks, and rotation enhance readability greater than any shade selection. Set solely the ticks your viewers wants. Use rotation when labels collide. Add small margins to cease markers from touching the body.

Earlier than the code: determine which ticks matter. For time or evenly spaced integers, present fewer ticks. For skewed information, take into account log scales and customized formatters.

import numpy as np
flights = sns.load_dataset("flights")  # columns: 12 months, month, passengers

ax = sns.lineplot(
    information=flights,
    x="12 months",
    y="passengers",
    estimator=None,
    errorbar=None,
    marker="o",
    dashes=False
)

ax.set_title("Airline passengers by 12 months")
ax.set_xlabel("Yr")
ax.set_ylabel("Passengers")

# Present a tick each 5 years
years = np.kind(flights["year"].distinctive())
ax.set_xticks(years[::5])

# Tidy the view
ax.margins(x=0.02, y=0.05)   # small padding inside axes
ax.set_ylim(0, None)         # begin at zero for clearer pattern

# Rotate if wanted
plt.xticks(rotation=0)
plt.tight_layout()
plt.present()

Extras you’ll be able to add when required

• Symmetric limits: ax.set_xlim(left, proper) and ax.set_ylim(backside, high) for honest comparisons
• Log scaling: ax.set_xscale("log") or ax.set_yscale("log") for lengthy tails
• Fewer ticks: ax.xaxis.set_major_locator(matplotlib.ticker.MaxNLocator(nbins=6))

// Annotations, traces, and spans

Annotations name out the rationale the plot exists. Use a brief label, a transparent arrow, and constant styling. Traces and spans mark thresholds or intervals that matter.

Earlier than the code: place annotations close to the information they seek advice from, however keep away from protecting factors. Think about using a semi-transparent span for ranges.

import matplotlib as mpl
suggestions = sns.load_dataset("suggestions")

ax = sns.regplot(
    information=suggestions,
    x="total_bill",
    y="tip",
    ci=None,
    scatter_kws=dict(s=28, alpha=0.6),
    line_kws=dict(linewidth=2)
)

ax.set_title("Tip vs whole invoice with callouts")
ax.set_xlabel("Complete invoice ($)")
ax.set_ylabel("Tip ($)")

# Threshold line for a tipping rule of thumb
ax.axhline(3, shade="#444", linewidth=1, linestyle="--")
ax.textual content(ax.get_xlim()[0], 3.1, "Reference: $3 tip", fontsize=9, shade="#444")

# Spotlight a invoice vary with a span
ax.axvspan(20, 40, shade="#fde68a", alpha=0.25, linewidth=0)  # comfortable spotlight

# Annotate a consultant level
pt = suggestions.loc[tips["total_bill"].between(20, 40)].iloc[0]
ax.annotate(
    "Instance verify",
    xy=(pt["total_bill"], pt["tip"]),
    xytext=(pt["total_bill"] + 10, pt["tip"] + 2),
    arrowprops=dict(
        arrowstyle="->",
        shade="#111",
        shrinkA=0,
        shrinkB=0,
        linewidth=1.2
    ),
    fontsize=9,
    bbox=dict(boxstyle="spherical,pad=0.2", fc="white", ec="#ddd", alpha=0.9)
)

plt.tight_layout()
plt.present()

Tips:

• Maintain annotations quick. The plot ought to nonetheless learn with out them
• Use axvline, axhline, axvspan, and axhspan for thresholds and ranges
• If labels overlap, regulate with small offsets or scale back font measurement, not by eradicating the annotation that carries that means

# Wrapping Up

 
You now have a whole baseline for quick, constant Seaborn work: pattern or combination when scale calls for it, management legends and axes with Matplotlib hooks, preserve colours steady throughout figures, and repair labels earlier than export. Mix these with the grid patterns and statistical plots from earlier sections and you may cowl most evaluation wants with out customized subplot code.

The place to be taught extra:

Shittu Olumide is a software program engineer and technical author enthusiastic about leveraging cutting-edge applied sciences to craft compelling narratives, with a eager eye for element and a knack for simplifying advanced ideas. You can too discover Shittu on Twitter.