Relationships, Matrix Charts & Styling in Seaborn: Complete Guide

Variables ke beech relationships visualize karna, correlation heatmaps banana, pair-wise analysis karna, aur charts ko professional themes se polish karna — sab ek final guide mein.

📑 Is Guide Mein 3 Sections Cover Honge:

• Section A — Relationship Charts: scatterplot(), lineplot(), regplot(), lmplot(), relplot()
• Section B — Matrix & Advanced: heatmap(), clustermap(), pairplot(), jointplot(), FacetGrid
• Section C — Styling & Theming: set_theme(), set_style(), set_palette(), set_context(), despine()

Section A — Relationship & Regression Charts

🔍 Kya Hai: Seaborn scatterplot() matplotlib scatter() ka supercharged version hai — hue (color), size (bubble), aur style (marker shape) se ek scatter plot mein 5 dimensions tak dikhane ka power deta hai. DataFrame directly accept karta hai.

💻 Professional Multi-Dimensional Version:

fig, ax = plt.subplots(figsize=(12, 8))

sns.scatterplot(data=df, x="Experience", y="Salary",
    hue="Department",              # color = department
    size="Rating",                 # dot size = rating (bubble chart!)
    style="Gender",                # marker shape = gender
    sizes=(20, 200),               # min/max dot size range
    palette="Set2",               # color palette
    alpha=0.7,                     # transparency
    edgecolor="#2c3e50",           # dot border
    linewidth=0.5,                 # border thickness
    markers=["o", "s"],             # circle for Male, square for Female
    ax=ax
)

ax.set_title("Experience vs Salary (Color=Dept, Size=Rating, Shape=Gender)",
    fontsize=14, fontweight="bold")
ax.set_xlabel("Experience (Years)", fontsize=12)
ax.set_ylabel("Salary (₹)", fontsize=12)
ax.legend(bbox_to_anchor=(1.05, 1), loc="upper left", fontsize=9)
sns.despine()
plt.savefig("scatter_multi.png", dpi=150, bbox_inches="tight")
plt.show()

📋 scatterplot() Parameters:

🔍 Kya Hai: Seaborn lineplot() automatically confidence band (shaded area) dikhata hai jab multiple observations same x-value par hain. Yeh matplotlib plot() mein manually karna padta hai lekin Seaborn mein built-in hai — time-series trends ke liye perfect.

💻 Professional Version:

fig, ax = plt.subplots(figsize=(12, 6))

sns.lineplot(data=df, x="Experience", y="Salary",
    hue="Department",              # separate line per department
    style="Gender",                # line style per gender
    palette="Set2",
    linewidth=2.5,
    markers=True,                  # show markers on data points
    dashes=True,                   # different dash styles per style group
    errorbar="ci",                 # "ci","sd","se","pi", None
    estimator="mean",              # aggregation function
    sort=True,                     # sort x values
    ax=ax
)

ax.set_title("Salary Trend by Experience (with CI Band)", fontsize=15, fontweight="bold")
ax.set_xlabel("Experience (Years)", fontsize=12)
ax.set_ylabel("Average Salary (₹)", fontsize=12)
ax.legend(bbox_to_anchor=(1.05, 1), fontsize=9)
sns.despine()
plt.savefig("lineplot_ci.png", dpi=150, bbox_inches="tight")
plt.show()

🔍 Kya Hai: regplot() scatter plot ke saath automatically regression line aur confidence interval band fit karta hai. Linear, polynomial, aur logistic regression sab built-in hain — ek line mein statistical relationship visualize ho jaata hai.

💻 Linear + Polynomial Regression:

fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 6))

# Linear Regression
sns.regplot(data=df, x="Experience", y="Salary",
    scatter_kws={"alpha": 0.3, "s": 20, "color": "#3498db"},
    line_kws={"color": "#e74c3c", "linewidth": 2.5},
    ci=95,                        # confidence interval (95%)
    order=1,                      # polynomial order (1=linear)
    ax=ax1
)
ax1.set_title("Linear Regression (order=1)", fontsize=14, fontweight="bold")

# Polynomial Regression
sns.regplot(data=df, x="Experience", y="Salary",
    scatter_kws={"alpha": 0.3, "s": 20, "color": "#2ecc71"},
    line_kws={"color": "#e74c3c", "linewidth": 2.5},
    ci=95,
    order=2,                      # polynomial degree 2 (quadratic)
    ax=ax2
)
ax2.set_title("Polynomial Regression (order=2)", fontsize=14, fontweight="bold")

for ax in [ax1, ax2]:
    sns.despine(ax=ax)
plt.tight_layout()
plt.savefig("regplot.png", dpi=150, bbox_inches="tight")
plt.show()

📋 regplot() Parameters:

🔍 Kya Hai: lmplot() regplot() ka figure-level version hai — col/row se automatic faceted regression panels bante hain. Har department ya gender ke liye separate regression dikhana ho toh lmplot() ek line mein kar deta hai.

# Faceted regression — separate panel per department
g = sns.lmplot(data=df, x="Experience", y="Salary",
    col="Department",              # separate column per department
    hue="Gender",                  # color by gender
    col_wrap=2,                   # 2 panels per row
    height=4,                     # panel height
    aspect=1.3,                   # width ratio
    palette=["#3498db", "#e74c3c"],
    scatter_kws={"alpha": 0.4, "s": 15},
    line_kws={"linewidth": 2},
    ci=95,
    order=1,
    robust=False
)

g.fig.suptitle("Salary vs Experience — Per Department", fontsize=16, fontweight="bold", y=1.03)
plt.savefig("lmplot.png", dpi=150, bbox_inches="tight")
plt.show()

🔍 Kya Hai: relplot() scatterplot() aur lineplot() dono ka figure-level version hai — kind="scatter" ya kind="line" se switch karo. col/row faceting built-in hai. Seaborn ka sabse versatile relational plotting function.

# Faceted scatter — Department panels + Gender color
g = sns.relplot(data=df, x="Experience", y="Salary",
    kind="scatter",                # "scatter" or "line"
    col="Department",
    hue="Gender",
    size="Rating",
    sizes=(20, 150),
    col_wrap=2,
    height=4,
    aspect=1.3,
    palette=["#3498db", "#e74c3c"],
    alpha=0.6
)

g.fig.suptitle("Multi-Dimensional Relationship Analysis", fontsize=16, fontweight="bold", y=1.03)
plt.savefig("relplot.png", dpi=150, bbox_inches="tight")
plt.show()

Section B — Matrix & Advanced Charts

🔍 Kya Hai: Seaborn heatmap() 2D numerical matrix ko color-coded grid mein visualize karta hai. Correlation matrix, confusion matrix, aur pivot table visualization — EDA ka sabse important single chart.

💻 Correlation Heatmap (Lower Triangle):

import numpy as np

corr = df.select_dtypes(include=["number"]).corr().round(2)
mask = np.triu(np.ones_like(corr, dtype=bool))  # upper triangle mask

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

sns.heatmap(corr,
    mask=mask,                     # hide upper triangle (duplicate)
    annot=True,                    # show values in cells
    fmt=".2f",                     # number format
    cmap="RdBu_r",                 # diverging colormap
    center=0,                      # center value
    vmin=-1, vmax=1,               # value range
    square=True,                   # square cells
    linewidths=0.5,               # cell border width
    linecolor="white",             # cell border color
    cbar_kws={"shrink": 0.8, "label": "Correlation"},
    annot_kws={"size": 11, "weight": "bold"},
    ax=ax
)

ax.set_title("Feature Correlation Matrix", fontsize=16, fontweight="bold", pad=15)
plt.savefig("heatmap_corr.png", dpi=150, bbox_inches="tight")
plt.show()

🔍 Kya Hai: clustermap() heatmap + hierarchical clustering combine karta hai — similar rows/columns ko automatically group karta hai aur dendrograms (tree diagrams) dikhata hai. Pattern discovery ke liye heatmap se better.

corr = df.select_dtypes(include=["number"]).corr()

g = sns.clustermap(corr,
    annot=True,                    # show values
    fmt=".2f",
    cmap="coolwarm",               # colormap
    center=0,
    linewidths=0.5,
    figsize=(10, 8),               # figure size
    method="ward",                # clustering method: "ward","complete","average","single"
    metric="euclidean",            # distance metric
    standard_scale=1,             # 0=row, 1=column standardization
    dendrogram_ratio=0.15,        # dendrogram size
    cbar_pos=(0.02, 0.8, 0.03, 0.15),
    tree_kws={"linewidths": 1.5}
)

g.fig.suptitle("Clustered Correlation Map", fontsize=16, fontweight="bold", y=1.02)
plt.savefig("clustermap.png", dpi=150, bbox_inches="tight")
plt.show()

🔍 Kya Hai: pairplot() har numerical column ka har doosre column ke saath scatter plot banata hai — diagonal par distribution. Ek call mein complete pairwise relationship matrix ban jaati hai. EDA ka most powerful single function.

g = sns.pairplot(df[["Salary", "Experience", "Age", "Rating", "Department"]],
    hue="Department",              # color by department
    palette="Set2",
    diag_kind="kde",              # diagonal: "hist","kde",None
    kind="scatter",                # off-diagonal: "scatter","kde","hist","reg"
    plot_kws={"alpha": 0.5, "s": 15},
    diag_kws={"alpha": 0.6, "linewidth": 2},
    height=2.5,                   # each panel size
    aspect=1,
    corner=True                    # lower triangle only (cleaner)
)

g.fig.suptitle("Pairwise Relationship Matrix", fontsize=16, fontweight="bold", y=1.02)
plt.savefig("pairplot.png", dpi=150, bbox_inches="tight")
plt.show()

📋 pairplot() Parameters:

🔍 Kya Hai: jointplot() center mein scatter/KDE dikhata hai aur margins (edges) par individual distributions — ek chart mein 3 views: relationship + x distribution + y distribution. Bivariate analysis ka complete picture.

# Scatter + Marginal Histograms
g = sns.jointplot(data=df, x="Experience", y="Salary",
    kind="scatter",                # "scatter","kde","hist","hex","reg","resid"
    hue="Gender",
    palette=["#3498db", "#e74c3c"],
    height=8,                     # figure size
    ratio=5,                      # center vs margin size ratio
    space=0.2,                    # gap between center and margins
    marginal_kws={"bins": 20, "alpha": 0.6},
    joint_kws={"alpha": 0.5, "s": 20}
)

g.fig.suptitle("Experience vs Salary — Joint Distribution", fontsize=15, fontweight="bold", y=1.02)
plt.savefig("jointplot.png", dpi=150, bbox_inches="tight")
plt.show()

# KDE Version — density contours
g = sns.jointplot(data=df, x="Experience", y="Salary",
    kind="kde", fill=True, cmap="YlOrRd", levels=15, height=8)
plt.savefig("jointplot_kde.png", dpi=150, bbox_inches="tight")
plt.show()

# Hex Version — large data density
g = sns.jointplot(data=df, x="Experience", y="Salary",
    kind="hex", cmap="Blues", height=8)
plt.savefig("jointplot_hex.png", dpi=150, bbox_inches="tight")
plt.show()

🔍 Kya Hai: FacetGrid sabse flexible multi-panel tool hai — koi bhi plotting function (matplotlib ya seaborn) ko faceted grid mein map kar sakte hain. displot, catplot, relplot sab internally FacetGrid use karte hain.

# Custom FacetGrid — any function mapped
g = sns.FacetGrid(df,
    col="Department",              # columns
    row="Gender",                  # rows
    hue="Gender",                  # color
    height=3.5,
    aspect=1.2,
    palette=["#3498db", "#e74c3c"],
    margin_titles=True            # show row/col titles on margins
)

# Map any plotting function
g.map_dataframe(sns.histplot, x="Salary", kde=True, bins=15, alpha=0.6)

# Add reference lines to all panels
g.map(lambda **kwargs: plt.axvline(df["Salary"].mean(), color="red", linestyle="--", alpha=0.5))

g.fig.suptitle("Salary Distribution — Department × Gender Grid", fontsize=16, fontweight="bold", y=1.03)
g.add_legend()
plt.savefig("facetgrid.png", dpi=150, bbox_inches="tight")
plt.show()

Section C — Styling & Theming

# ===== set_theme() — All-in-one global configuration =====
sns.set_theme(
    style="whitegrid",             # background style
    context="notebook",            # sizing context
    palette="Set2",               # default colors
    font="Arial",                  # font family
    font_scale=1.1,               # font size multiplier
    rc={                            # custom rcParams
        "figure.figsize": (10, 6),
        "axes.titleweight": "bold",
        "axes.labelweight": "bold"
    }
)

# ===== 5 BUILT-IN STYLES =====
sns.set_style("whitegrid")      # ✅ White bg + grid (MOST POPULAR)
sns.set_style("darkgrid")       # Gray bg + grid (seaborn default)
sns.set_style("white")          # Clean white — minimal
sns.set_style("dark")           # Dark bg — no grid
sns.set_style("ticks")          # White bg + tick marks (scientific)

# Custom overrides
sns.set_style("whitegrid", {
    "grid.linestyle": "--",
    "grid.alpha": 0.3,
    "axes.edgecolor": "#cccccc"
})

# Temporary style
with sns.axes_style("dark"):
    fig, ax = plt.subplots()
    sns.histplot(data=df, x="Salary", ax=ax)
    plt.show()
# Original style restored automatically!

# ===== BUILT-IN PALETTES =====

# Qualitative (distinct categories)
sns.set_palette("Set1")          # bold, distinct
sns.set_palette("Set2")          # pastel, professional ✅
sns.set_palette("Set3")          # light pastel
sns.set_palette("Pastel1")       # very light
sns.set_palette("tab10")         # matplotlib default 10 colors

# Sequential (gradient)
sns.set_palette("Blues")          # light to dark blue
sns.set_palette("YlOrRd")        # yellow to red (heatmaps)

# Diverging
sns.set_palette("RdBu")          # red to blue (correlations)
sns.set_palette("coolwarm")      # cool blue to warm red

# ===== CUSTOM PALETTE =====
my_colors = ["#3498db", "#2ecc71", "#e74c3c", "#f39c12", "#9b59b6"]
sns.set_palette(my_colors)

# Generate palette programmatically
pal = sns.color_palette("husl", 8)         # 8 evenly spaced hues
pal = sns.color_palette("light:#3498db", 6) # light to dark blue gradient
pal = sns.color_palette("blend:#3498db,#e74c3c", 5)  # blue to red blend

# Preview palette
sns.palplot(sns.color_palette("Set2"))
plt.title("Set2 Palette")
plt.savefig("palette_preview.png", dpi=150, bbox_inches="tight")
plt.show()

# ===== 4 CONTEXT PRESETS =====
sns.set_context("paper")         # smallest — research papers
sns.set_context("notebook")      # medium — Jupyter notebooks ✅ (default)
sns.set_context("talk")          # large — presentations/slides
sns.set_context("poster")        # largest — posters/billboards

# Custom font scale
sns.set_context("notebook", font_scale=1.3)

# Custom overrides
sns.set_context("talk", rc={
    "lines.linewidth": 3,
    "axes.titlesize": 18
})

# ===== DESPINE OPTIONS =====

# Default — remove top & right spines
sns.despine()

# Custom — choose which to remove
sns.despine(
    top=True,                     # remove top spine
    right=True,                   # remove right spine
    left=False,                    # keep left spine
    bottom=False,                  # keep bottom spine
    offset=10,                    # move spines away from data (pixels)
    trim=True                     # trim spines to data range
)

# Specific axes
fig, (ax1, ax2) = plt.subplots(1, 2)
sns.despine(ax=ax1)              # only despine ax1
sns.despine(ax=ax2, left=True)   # remove left spine too from ax2

# ===== COPY-PASTE THIS AT TOP OF EVERY NOTEBOOK =====

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

# Professional theme setup
sns.set_theme(
    style="whitegrid",
    context="notebook",
    palette="Set2",
    font_scale=1.1,
    rc={
        "figure.figsize": (10, 6),
        "figure.dpi": 100,
        "savefig.dpi": 150,
        "axes.titleweight": "bold",
        "axes.labelweight": "bold",
        "axes.titlesize": 14,
        "axes.labelsize": 12,
        "axes.spines.top": False,
        "axes.spines.right": False
    }
)

print("✅ Professional visualization setup loaded!")

🎉 Seaborn Masterclass Complete!

Congratulations! Aapne 3 comprehensive parts mein 15 Seaborn functions cover kiye hain — Distribution se Matrix Charts tak, Styling se Professional Templates tak. Ab aap publication-quality statistical visualizations confidently create kar sakte hain.

Happy Visualizing! 📊🚀