Amounts and Distribution

Python

Seaborn

Distribution

Use Seaborn to create visualizations for amounts & distributions: barplots, point plots, histograms, box plots, etc.

Author

Pinn Xu

Published

October 1, 2024

9 min read · 1755 EN words · 0 中文字

Environment and Versions

python==3.12
pandas==2.2.2
matplotlib==3.9.x
seaborn==0.13.2

1 Seaborn and Matplotlib

Note: If you are using a local environment such as Jupyter or VS Code, install Seaborn first. If you are using Google Colab, Seaborn is already available and does not need to be installed manually.

pip install seaborn

import seaborn as sns
import matplotlib.pyplot as plt

matplotlib.pyplot is usually abbreviated as plt so plotting functions can be called more conveniently. Using the full path is not recommended:

import matplotlib.pyplot
matplotlib.pyplot.plot(...)
matplotlib.pyplot.show()

Recommended style:

import matplotlib.pyplot as plt  # Import with the alias 'plt' for later use
plt.plot(...)
plt.show()

mpl: an alias for matplotlib, commonly used for global configuration, lower-level properties, and advanced submodules such as colormaps and 3D plotting.
plt: an alias for matplotlib.pyplot, the main everyday interface for quickly creating and managing figures.

Reference: matplotlib.pyplot tutorial

df = sns.load_dataset('penguins')  # Load the penguins example dataset
df.info()  # View column names, data types, and non-null counts

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 344 entries, 0 to 343
Data columns (total 7 columns):
 #   Column             Non-Null Count  Dtype  
---  ------             --------------  -----  
 0   species            344 non-null    object 
 1   island             344 non-null    object 
 2   bill_length_mm     342 non-null    float64
 3   bill_depth_mm      342 non-null    float64
 4   flipper_length_mm  342 non-null    float64
 5   body_mass_g        342 non-null    float64
 6   sex                333 non-null    object 
dtypes: float64(4), object(3)
memory usage: 18.9+ KB

Float Data

Floating-point columns such as length and mass are continuous numerical variables. Variables with a true zero, where 0 means “none”, are ratio data, such as length and mass. Variables without a true zero are interval data, such as temperature.

Object Data

Species names and island names are nominal data. Sex is ordinal data.

df.head()  # View the first five rows

	species	island	bill_length_mm	bill_depth_mm	flipper_length_mm	body_mass_g	sex
0	Adelie	Torgersen	39.1	18.7	181.0	3750.0	Male
1	Adelie	Torgersen	39.5	17.4	186.0	3800.0	Female
2	Adelie	Torgersen	40.3	18.0	195.0	3250.0	Female
3	Adelie	Torgersen	NaN	NaN	NaN	NaN	NaN
4	Adelie	Torgersen	36.7	19.3	193.0	3450.0	Female

df.tail()  # View the last five rows

	species	island	bill_length_mm	bill_depth_mm	flipper_length_mm	body_mass_g	sex
339	Gentoo	Biscoe	NaN	NaN	NaN	NaN	NaN
340	Gentoo	Biscoe	46.8	14.3	215.0	4850.0	Female
341	Gentoo	Biscoe	50.4	15.7	222.0	5750.0	Male
342	Gentoo	Biscoe	45.2	14.8	212.0	5200.0	Female
343	Gentoo	Biscoe	49.9	16.1	213.0	5400.0	Male

df.sample(5)  # Randomly sample five rows

	species	island	bill_length_mm	bill_depth_mm	flipper_length_mm	body_mass_g	sex
123	Adelie	Torgersen	41.4	18.5	202.0	3875.0	Male
166	Chinstrap	Dream	45.9	17.1	190.0	3575.0	Female
240	Gentoo	Biscoe	50.2	14.3	218.0	5700.0	Male
7	Adelie	Torgersen	39.2	19.6	195.0	4675.0	Male
19	Adelie	Torgersen	46.0	21.5	194.0	4200.0	Male

Note: Function vs. Method

df.info(), df.head(), df.tail(), and df.sample() are all methods of a DataFrame object. They must be called from a specific object and cannot be used independently.

pd.read_csv() is a predefined Pandas function and can be called directly:

df = pd.read_csv('file_path')

By contrast, df.sample() is a method that operates on the df object:

df.sample(4)

	species	island	bill_length_mm	bill_depth_mm	flipper_length_mm	body_mass_g	sex
227	Gentoo	Biscoe	46.7	15.3	219.0	5200.0	Male
120	Adelie	Torgersen	36.2	17.2	187.0	3150.0	Female
272	Gentoo	Biscoe	45.1	14.4	210.0	4400.0	Female
62	Adelie	Biscoe	37.6	17.0	185.0	3600.0	Female

2 Drawing Bar Plots

2.1 Basic Bar Plots with Seaborn

2.1.1 Basic Bar Plot

# Basic bar plot: the x-axis is categorical, and the y-axis shows the mean by default.
# The black error bars show 95% confidence intervals.
sns.barplot(data=df, x='island', y='body_mass_g')
plt.show()

# estimator='sum': use the sum as the bar height instead of the default mean
sns.barplot(data=df, x='island', y='body_mass_g', estimator='sum')
plt.show()

Reference: sns.barplot() documentation (estimator parameter)

sns.__version__  # View the current Seaborn version. Since 0.13, the default is a single color to avoid meaningless color distinctions.

'0.13.2'

# Check: the mean body mass for Biscoe Island is the bar height shown in the chart.
df.loc[df['island'] == 'Biscoe', 'body_mass_g'].mean()

4716.017964071856

2.1.2 Count Plot

# Count plot: automatically counts the number of observations in each category
sns.countplot(data=df, x='species')
plt.show()

Reference: sns.countplot() documentation

2.2 Grouping Bars and Removing Error Bars

Use the hue parameter to group by category, and set errorbar=None to remove error bars.

# Group by sex with hue, and remove error bars with errorbar=None
sns.barplot(data=df, x='island', y='body_mass_g', errorbar=None, hue='sex')
plt.show()

Reference: sns.barplot() documentation (errorbar and hue parameters)

2.3 Rotating Tick Labels

When x-axis labels are too long, use plt.xticks() to rotate them:

sns.barplot(data=df, x='island', y='body_mass_g', errorbar=None, hue='sex')

# Rotate labels by 45 degrees. ha='right' aligns the right edge of each label with the tick.
plt.xticks(rotation=45, ha='right')
plt.show()

Reference: plt.xticks() documentation (kwargs section)

Wilke (2019, chap. 6.1) suggests that when labels are dense, a horizontal layout is easier to read than rotated labels:

# Swap the x and y axes: a horizontal layout is easier to read when labels are dense.
sns.barplot(data=df, y='island', x='body_mass_g', errorbar=None, hue='sex')
plt.show()

The same idea applies to count plots:

sns.countplot(data=df, y='species', hue='sex')
plt.show()

2.4 Setting the Global Theme

# Set the global plotting theme. It applies to all subsequent Seaborn charts.
# Use sns.reset_orig() to restore the default theme.
sns.set_theme(font_scale=1.2, style='darkgrid')

sns.set_theme() sets the global plotting theme for all later Seaborn charts. Call sns.reset_orig() to restore the default theme.

# Define a chart object named ax
ax = sns.barplot(
    data=df, x='body_mass_g', y='island',
    errorbar=None, hue='sex',
)

# Set the title and axis labels
ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island')
plt.show()

After assigning a Seaborn chart to the variable ax, you can call Matplotlib methods on it for further customization, similar to assigning a DataFrame to df and then calling its methods.

Note the difference between these two:

sns.set() / sns.set_theme(): Seaborn global theme-setting functions. They are equivalent, but sns.set_theme() is preferred.
ax.set(): a Matplotlib method for a specific chart object, used to set titles, axis labels, and related elements.

References:
sns.set() documentation
sns.set_theme() documentation
ax.set() documentation

2.5 Adjusting the Legend

Whether you use errorbar=None or errorbar=('ci', 0), the legend may cover the plot, so you may need to adjust its position manually with sns.move_legend().

# errorbar=None: the legend is automatically placed on the right
ax = sns.barplot(data=df, y='island', x='body_mass_g', errorbar=None, hue='sex')
ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island')
plt.show()

# errorbar=('ci', 0): the legend is automatically moved away from the error-bar area
ax = sns.barplot(
    data=df, y='island', x='body_mass_g', 
    errorbar=('ci', 0), hue='sex',
)
ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island')
plt.show()

ax = sns.barplot(data=df, y='island', x='body_mass_g', errorbar=None, hue='sex')
ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island')

# Combine bbox_to_anchor with loc: move the legend's loc reference point to (0.5, -0.2)
sns.move_legend(ax, bbox_to_anchor=(0.5, -0.2), loc='upper center', ncols=2)
plt.show()

ax = sns.barplot(data=df, y='island', x='body_mass_g', errorbar=None, hue='sex')
ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island')
sns.move_legend(ax, loc='lower right')  # Set loc only: move the legend to the lower right of the plot
plt.show()

ax = sns.barplot(data=df, y='island', x='body_mass_g', errorbar=None, hue='sex')
ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island')

# Place the upper-left corner of the legend at (1, 1) relative to the axes
sns.move_legend(ax, loc='upper left', bbox_to_anchor=(1, 1))
plt.show()

References:
sns.move_legend() documentation (obj and loc parameters)
matplotlib.axes.Axes.legend() documentation (loc, ncols, and bbox_to_anchor parameters)

2.6 Changing Colors

Use the palette parameter to change chart colors. It can take the following kinds of values:

Seaborn predefined palette names (default: 'deep'): 'deep', 'muted', 'bright', 'pastel', 'dark', 'colorblind'
Matplotlib colormap names: for example, 'viridis', 'plasma', 'coolwarm', 'Blues', 'Reds'
HLS/HUSL color models: 'hls', 'husl'
Cubehelix schemes: 'ch:<args>', for example 'ch:s=0.5,r=-0.5' (colorblind-friendly and suitable for grayscale printing)
Gradients: 'light:<color>', 'dark:<color>'
Blends: 'blend:<color1>,<color2>'
Lists of color codes: hexadecimal codes such as ['#FF5733', '#33FF57'], RGB tuples, or Matplotlib color names

# Default palette
ax = sns.barplot(data=df, y='island', x='body_mass_g', errorbar=None, hue='sex')
ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island')
sns.move_legend(ax, bbox_to_anchor=(1, 1), loc='upper left')
plt.show()

# palette='flare': use a predefined Seaborn palette
ax = sns.barplot(data=df, y='island', x='body_mass_g', errorbar=None, hue='sex', palette='flare')
ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island')
sns.move_legend(ax, bbox_to_anchor=(1, 1), loc='upper left')
plt.show()

# Pass a list of Matplotlib color names to palette
ax = sns.barplot(data=df, y='island', x='body_mass_g', errorbar=None, hue='sex', palette=['blue', 'orange'])
ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island')
sns.move_legend(ax, bbox_to_anchor=(1, 1), loc='upper left')
plt.show()

# Pass a list of hexadecimal color codes to palette
ax = sns.barplot(
    data=df, y='island', x='body_mass_g', 
    errorbar=None, hue='sex', palette=['#a1c9f4', '#8de5a1'],
)
ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island')
sns.move_legend(ax, bbox_to_anchor=(1, 1), loc='upper left')
plt.show()

References:
Seaborn sns.set_theme() documentation (palette parameter)
Seaborn sns.color_palette() documentation (list of predefined palette names)
Matplotlib named colors
Matplotlib colormaps
w3schools color picker

Extension: Counting Penguins on Each Island

The following two approaches are equivalent and can be used later to sort bar plots by totals.

Method 1: df.groupby().size()

# Group by island and count the number of rows in each group
df.groupby('island').size()

island
Biscoe       168
Dream        124
Torgersen     52
dtype: int64

Method 2: df.value_counts()

# Count the occurrences of each value in the island column
df.value_counts('island')

island
Biscoe       168
Dream        124
Torgersen     52
Name: count, dtype: int64

3 Drawing Other Plots

3.1 Point Plot

The y-axis of a bar plot must start at 0 to ensure that the visual proportions match the data proportions, following Wilke’s Principle of Proportional Ink (Wilke 2019, chap. 17). Point plots do not have this restriction: the y-axis can start from the data range, and the data-ink ratio is higher because there is less visual clutter.

When the x-axis is nominal data, such as island names, the points do not need to be connected with lines.

# linestyle='none': do not connect points when the x-axis is categorical.
# The vertical segments around points are error bars.
sns.pointplot(data=df, x='island', y='body_mass_g', hue='sex', linestyle='none')
plt.show()

Reference: sns.pointplot() documentation

3.2 Histogram

The x-axis of a bar plot represents discrete categories (nominal or ordinal data). The x-axis of a histogram represents continuous numerical intervals (ratio data).

sns.histplot(data=df, x='flipper_length_mm')
plt.show()

# Group with hue: plot penguins from each island separately and overlay the distributions
sns.histplot(data=df, x='flipper_length_mm', hue='island')
plt.show()

# Use a consistent y-axis range to compare the plot with and without hue grouping
sns.histplot(data=df, x='flipper_length_mm')
plt.show()

sns.histplot(data=df, x='flipper_length_mm', hue='island')
plt.ylim(0, 80)
plt.show()

sns.histplot(
    data=df, x='flipper_length_mm',
    hue='island',
    stat='density',     # The total area of all bins is 1 (probability density)
    common_norm=False   # Normalize each group separately. If True, all groups share one total area of 1.
)
plt.show()

3.2.1 The `stat` Parameter

This parameter controls the statistic shown by the histogram. Available values include:

'count': bin height = number of data points in the bin.
'frequency': bin height = count / bin width, or frequency per unit width. This is useful when bin widths are unequal.
'probability' / 'proportion': bin height = count / total count, so all bin heights sum to 1.
'percent': bin height = (count / total count) * 100, so all bin heights sum to 100.
'density': bin height = frequency / bin width, so the total area of all bins is 1. As bin width approaches 0, this becomes a smooth density curve.

# stat='count': bar height = number of data points in each bin
ax = sns.histplot(data=df, x='flipper_length_mm', stat='count')

# Iterate over each bar (patch), get its height, and show a value label above it
for p in ax.patches:
    # Get the bar height, which is the frequency in the interval
    height = p.get_height()
    # Add text
    ax.text(
        p.get_x() + p.get_width() / 2,  # x coordinate: center of the bar
        height, # y coordinate: bar height
        int(height), # text: integer frequency
        ha='center', va='bottom', # horizontal/vertical alignment
    )
plt.show()

# stat='frequency': bar height = count / bin width
ax = sns.histplot(data=df, x='flipper_length_mm', stat='frequency')
for p in ax.patches:
    height = p.get_height()
    ax.text(
        p.get_x() + p.get_width() / 2, 
        height, int(height), 
        ha='center', va='bottom',
    )
plt.show()

# stat='probability': bar height = count / total count, and all bar heights sum to 1
ax = sns.histplot(data=df, x='flipper_length_mm', stat='probability')
for p in ax.patches:
    height = p.get_height()
    ax.text(
        p.get_x() + p.get_width() / 2, 
        height, f'{height:.2f}', # Keep two decimal places
        ha='center', va='bottom',
    )
plt.show()

# stat='percent': bar height = (count / total count) * 100, and all bar heights sum to 100
ax = sns.histplot(data=df, x='flipper_length_mm', stat='percent')
for p in ax.patches:
    height = p.get_height()
    ax.text(
        p.get_x() + p.get_width() / 2, 
        height, int(height), 
        ha='center', va='bottom',
    )
plt.show()

# stat='density': bar height = frequency / bin width, and the total bin area is 1
ax = sns.histplot(data=df, x='flipper_length_mm', stat='density')
for p in ax.patches:
    height = p.get_height()
    ax.text(
        p.get_x() + p.get_width() / 2, 
        height, f'{height:.3f}', 
        ha='center', va='bottom',
    )
plt.show()

3.2.2 The `common_norm` Parameter

This boolean parameter only applies when stat is 'probability', 'proportion', 'percent', or 'density'. Its default value is True:

common_norm=True: normalize using the entire dataset, so bin heights can be compared across groups.
common_norm=False: normalize each subgroup independently, so bin heights are only meaningful within each group.

sns.histplot(
    data=df, x='flipper_length_mm', hue='island', 
    stat='density', common_norm=True, # Overall normalization: the total bin area across all groups is 1
)
plt.show()

sns.histplot(
    data=df, x='flipper_length_mm', hue='island',
    stat='density', common_norm=False, # Separate normalization: each group's bin area sums to 1
)
plt.show()

When overlapping groups are difficult to read, use data=df[df['column']==value] to filter the data and plot each group separately:

# Plot by island to avoid overlapping groups and inspect each island's distribution separately
ax = sns.histplot(
    data=df[df['island'] == 'Torgersen'], 
    x='flipper_length_mm', stat='density'
)
ax.set(title='Torgersen Island', xlabel='Flipper Length (mm)')
plt.show()

ax = sns.histplot(
    data=df[df['island'] == 'Biscoe'], 
    x='flipper_length_mm', stat='density'
)
ax.set(title='Biscoe Island', xlabel='Flipper Length (mm)')
plt.show()

ax = sns.histplot(
    data=df[df['island'] == 'Dream'], 
    x='flipper_length_mm', stat='density'
)
ax.set(title='Dream Island', xlabel='Flipper Length (mm)')
plt.show()

Reference: sns.histplot() documentation (hue, stat, and common_norm parameters)

3.3 Box Plot

A box plot summarizes a distribution with five key values: min, Q1, Q2 (median), Q3, and max. It also marks outliers beyond the whisker range.

Box: from Q1 to Q3, containing the middle 50% of the data.
Line inside the box: the median (Q2).
Whiskers: extend outward from Q1 and Q3, by default to the extreme values within 1.5 * IQR.
Outliers: points beyond the whisker range, marked separately.

Box Plot vs. Histogram

Box plots are useful for quickly identifying a distribution summary and outliers. Histograms are better for seeing the frequency distribution within specific intervals.

sns.boxplot(data=df, x='flipper_length_mm', y='island', hue='sex')
plt.show()

sns.boxplot(
    data=df, x='flipper_length_mm', y='island', hue='sex',
    fliersize=8,  # fliersize: size of outlier points (default: 5)
)
plt.show()

sns.boxplot(
    data=df, x='flipper_length_mm', y='island', hue='sex', fliersize=8,
    whis=2,  # whisker: extend to Q1/Q3 +/- 2*IQR (default: 1.5); larger values reduce outlier points
)
plt.show()

Reference: sns.boxplot() documentation (fliersize and whis parameters)

3.4 Violin Plot

A violin plot combines features of a box plot, such as medians and quantiles, with a density curve that shows the smoothed distribution. When the dataset is small, the smoothed curve may be misleading, so a histogram or box plot may be preferable.

sns.violinplot(data=df, y='island', x='flipper_length_mm', hue='sex')
plt.show()

# split=True: combine the two hue groups in the same category into one violin for easier distribution comparison
sns.violinplot(
    data=df, y='island', x='flipper_length_mm', hue='sex', 
    split=True, # Combine two hue groups into one violin for easier comparison
)
plt.show()

Reference: sns.violinplot() documentation (split parameter)

3.5 Strip Plot and Swarm Plot

Strip plots draw every data point, with each dot representing one observation. Points are randomly jittered along the categorical axis to reduce overlap. Swarm plots automatically arrange overlapping data points and are easier to read.

Note that both plots show a univariate distribution, meaning how densely one variable appears across its range. This is different from a scatter plot, which shows the relationship between two variables.

Comparison	Strip Plot	Swarm Plot
Basic idea	Randomly jitter data points along the categorical axis to avoid overlap	Precisely arrange points with an algorithm to avoid overlap and show the distribution
Point placement	Random jitter; overlap may still occur	Non-random; automatically adjusted by density to avoid overlap
Readability	Clear for small datasets; can overlap heavily with large datasets	Clearer for showing distribution shape in small to medium datasets
Computational cost	Lower and faster to draw	Higher and slower for large datasets
Suitable uses	Quick visualization, small datasets, situations where exact distribution shape is less important	Showing distribution details for medium-sized datasets
Distribution detail	Moderate; true density can be hard to see	Stronger; data density is easier to read visually
Parameter control	Jitter can be controlled with `jitter`	No jitter needed; the algorithm adjusts placement automatically
Example function	`sns.stripplot()`	`sns.swarmplot()`

sns.stripplot(data=df, x='flipper_length_mm', y='island', hue='sex')
plt.show()

sns.swarmplot(data=df, x='flipper_length_mm', y='island', hue='sex')
plt.show()

References:
sns.stripplot() documentation
sns.swarmplot() documentation

Reference Links

General links:
Python 3 documentation (for built-in functions and related usage)
Pandas documentation (DataFrame)
Matplotlib documentation (plt)
Seaborn documentation (barplot)

sns.barplot() documentation (drawing bar plots with Seaborn)
sns.countplot() documentation (drawing count plots with Seaborn)

Changing chart themes:
plt.xticks() documentation (the kwargs section explains how to rotate tick labels)
sns.set() documentation (equivalent to sns.set_theme())
sns.set_theme() documentation (set the global Seaborn plotting theme)
ax.set() documentation (set titles, axis labels, and related elements for a specific Matplotlib axes object)

Changing legends:
sns.move_legend() documentation (Seaborn function for moving legends; see the obj and loc parameters)
matplotlib.axes.Axes.legend() documentation (Matplotlib legend function; its options also apply to sns.move_legend(); see loc, ncols, and bbox_to_anchor)

Changing chart colors:
Seaborn sns.color_palette() documentation (predefined palette names for the palette parameter)
Matplotlib named colors
Matplotlib colormaps
Matplotlib colormap guide
w3schools color picker (view hexadecimal codes for colors)

Other charts:
sns.pointplot() documentation (drawing point plots)
sns.histplot() documentation (drawing histograms; see hue, stat, and common_norm)
sns.boxplot() documentation (drawing box plots; see fliersize and whis)
sns.violinplot() documentation (drawing violin plots)
sns.stripplot() documentation (drawing distribution scatter plots)
sns.swarmplot() documentation (drawing clustered distribution scatter plots)

Textbook: Fundamentals of Data Visualization (Wilke, 2019):
6.1 Bar Plot (rotated tick labels are harder to read; a horizontal layout is preferable)
17. The Principle of Proportional Ink (the proportions of graphical marks should match the true proportions in the data; for example, bars in a bar plot should start from 0)

⇧ Back to top

--- title: 'Amounts and Distribution' author: 'Pinn Xu' date: '2024-10-01' # categories: ["Python", "Seaborn", "Matplotlib", "Visualization", "Distribution", "Histogram", "Bar Plot", "Point Plot", "Box Plot"] categories: ["Python", "Seaborn", "Distribution"] description: "Use Seaborn to create visualizations for amounts & distributions: barplots, point plots, histograms, box plots, etc." code-fold: false order: 2 aliases: - ../amounts-and-distribution/ --- ```{python} #| eval: false #| code-fold: true #| code-summary: "Environment and Versions" python==3.12 pandas==2.2.2 matplotlib==3.9.x seaborn==0.13.2 ``` # Seaborn and Matplotlib > **Note:** If you are using a local environment such as Jupyter or VS Code, install Seaborn first. If you are using Google Colab, Seaborn is already available and does not need to be installed manually. ```{python} #| eval: false pip install seaborn ``` ```{python} import seaborn as sns import matplotlib.pyplot as plt ``` `matplotlib.pyplot` is usually abbreviated as `plt` so plotting functions can be called more conveniently. Using the full path is not recommended: ```{python} #| eval: false import matplotlib.pyplot matplotlib.pyplot.plot(...) matplotlib.pyplot.show() ``` Recommended style: ```{python} #| eval: false import matplotlib.pyplot as plt # Import with the alias 'plt' for later use plt.plot(...) plt.show() ``` - `mpl`: an alias for `matplotlib`, commonly used for global configuration, lower-level properties, and advanced submodules such as colormaps and 3D plotting. - `plt`: an alias for `matplotlib.pyplot`, the main everyday interface for quickly creating and managing figures. **Reference: [matplotlib.pyplot tutorial](https://matplotlib.org/stable/api/pyplot_summary.html)** ```{python} df = sns.load_dataset('penguins') # Load the penguins example dataset df.info() # View column names, data types, and non-null counts ``` **Float Data** Floating-point columns such as length and mass are continuous numerical variables. Variables with a true zero, where 0 means "none", are ratio data, such as length and mass. Variables without a true zero are interval data, such as temperature. **Object Data** Species names and island names are nominal data. Sex is ordinal data. ```{python} df.head() # View the first five rows ``` ```{python} df.tail() # View the last five rows ``` ```{python} df.sample(5) # Randomly sample five rows ``` **Note: Function vs. Method** `df.info()`, `df.head()`, `df.tail()`, and `df.sample()` are all **methods** of a DataFrame object. They must be called from a specific object and cannot be used independently. `pd.read_csv()` is a predefined Pandas **function** and can be called directly: ```{python} #| eval: false df = pd.read_csv('file_path') ``` By contrast, `df.sample()` is a method that operates on the `df` object: ```{python} df.sample(4) ``` # Drawing Bar Plots ## Basic Bar Plots with Seaborn ### Basic Bar Plot ```{python} # Basic bar plot: the x-axis is categorical, and the y-axis shows the mean by default. # The black error bars show 95% confidence intervals. sns.barplot(data=df, x='island', y='body_mass_g') plt.show() ``` ```{python} # estimator='sum': use the sum as the bar height instead of the default mean sns.barplot(data=df, x='island', y='body_mass_g', estimator='sum') plt.show() ``` **Reference: [`sns.barplot()` documentation](https://seaborn.pydata.org/generated/seaborn.barplot.html)** (`estimator` parameter) ```{python} sns.__version__ # View the current Seaborn version. Since 0.13, the default is a single color to avoid meaningless color distinctions. ``` ```{python} # Check: the mean body mass for Biscoe Island is the bar height shown in the chart. df.loc[df['island'] == 'Biscoe', 'body_mass_g'].mean() ``` ### Count Plot ```{python} # Count plot: automatically counts the number of observations in each category sns.countplot(data=df, x='species') plt.show() ``` **Reference: [`sns.countplot()` documentation](https://seaborn.pydata.org/generated/seaborn.countplot.html)** ## Grouping Bars and Removing Error Bars Use the `hue` parameter to group by category, and set `errorbar=None` to remove error bars. ```{python} # Group by sex with hue, and remove error bars with errorbar=None sns.barplot(data=df, x='island', y='body_mass_g', errorbar=None, hue='sex') plt.show() ``` **Reference: [`sns.barplot()` documentation](https://seaborn.pydata.org/generated/seaborn.barplot.html)** (`errorbar` and `hue` parameters) ## Rotating Tick Labels When x-axis labels are too long, use `plt.xticks()` to rotate them: ```{python} sns.barplot(data=df, x='island', y='body_mass_g', errorbar=None, hue='sex') # Rotate labels by 45 degrees. ha='right' aligns the right edge of each label with the tick. plt.xticks(rotation=45, ha='right') plt.show() ``` **Reference: [`plt.xticks()` documentation](https://matplotlib.org/stable/api/_as_gen/matplotlib.pyplot.xticks.html)** (`kwargs` section) [Wilke (2019, chap. 6.1)](https://clauswilke.com/dataviz/visualizing-amounts.html#bar-plots) suggests that when labels are dense, a horizontal layout is easier to read than rotated labels: ```{python} # Swap the x and y axes: a horizontal layout is easier to read when labels are dense. sns.barplot(data=df, y='island', x='body_mass_g', errorbar=None, hue='sex') plt.show() ``` The same idea applies to count plots: ```{python} sns.countplot(data=df, y='species', hue='sex') plt.show() ``` ## Setting the Global Theme ```{python} # Set the global plotting theme. It applies to all subsequent Seaborn charts. # Use sns.reset_orig() to restore the default theme. sns.set_theme(font_scale=1.2, style='darkgrid') ``` `sns.set_theme()` sets the global plotting theme for all later Seaborn charts. Call `sns.reset_orig()` to restore the default theme. ```{python} # Define a chart object named ax ax = sns.barplot( data=df, x='body_mass_g', y='island', errorbar=None, hue='sex', ) # Set the title and axis labels ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island') plt.show() ``` After assigning a Seaborn chart to the variable `ax`, you can call Matplotlib methods on it for further customization, similar to assigning a DataFrame to `df` and then calling its methods. **Note the difference between these two:** - `sns.set()` / `sns.set_theme()`: Seaborn **global** theme-setting functions. They are equivalent, but `sns.set_theme()` is preferred. - `ax.set()`: a Matplotlib method for a **specific chart object**, used to set titles, axis labels, and related elements. **References:**\ [`sns.set()` documentation](https://seaborn.pydata.org/generated/seaborn.set.html)\ [`sns.set_theme()` documentation](https://seaborn.pydata.org/generated/seaborn.set_theme.html#seaborn.set_theme)\ [`ax.set()` documentation](https://matplotlib.org/stable/api/_as_gen/matplotlib.axes.Axes.set.html) ## Adjusting the Legend Whether you use `errorbar=None` or `errorbar=('ci', 0)`, the legend may cover the plot, so you may need to adjust its position manually with `sns.move_legend()`. ```{python} # errorbar=None: the legend is automatically placed on the right ax = sns.barplot(data=df, y='island', x='body_mass_g', errorbar=None, hue='sex') ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island') plt.show() ``` ```{python} # errorbar=('ci', 0): the legend is automatically moved away from the error-bar area ax = sns.barplot( data=df, y='island', x='body_mass_g', errorbar=('ci', 0), hue='sex', ) ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island') plt.show() ``` ```{python} ax = sns.barplot(data=df, y='island', x='body_mass_g', errorbar=None, hue='sex') ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island') # Combine bbox_to_anchor with loc: move the legend's loc reference point to (0.5, -0.2) sns.move_legend(ax, bbox_to_anchor=(0.5, -0.2), loc='upper center', ncols=2) plt.show() ``` ```{python} ax = sns.barplot(data=df, y='island', x='body_mass_g', errorbar=None, hue='sex') ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island') sns.move_legend(ax, loc='lower right') # Set loc only: move the legend to the lower right of the plot plt.show() ``` ```{python} ax = sns.barplot(data=df, y='island', x='body_mass_g', errorbar=None, hue='sex') ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island') # Place the upper-left corner of the legend at (1, 1) relative to the axes sns.move_legend(ax, loc='upper left', bbox_to_anchor=(1, 1)) plt.show() ``` **References:**\ [`sns.move_legend()` documentation](https://seaborn.pydata.org/generated/seaborn.move_legend.html) (`obj` and `loc` parameters)\ [`matplotlib.axes.Axes.legend()` documentation](https://matplotlib.org/stable/api/_as_gen/matplotlib.axes.Axes.legend.html) (`loc`, `ncols`, and `bbox_to_anchor` parameters) ## Changing Colors Use the `palette` parameter to change chart colors. It can take the following kinds of values: 1. **Seaborn predefined palette names** (default: `'deep'`): `'deep'`, `'muted'`, `'bright'`, `'pastel'`, `'dark'`, `'colorblind'` 2. **Matplotlib colormap names**: for example, `'viridis'`, `'plasma'`, `'coolwarm'`, `'Blues'`, `'Reds'` 3. **HLS/HUSL color models**: `'hls'`, `'husl'` 4. **Cubehelix schemes**: `'ch:<args>'`, for example `'ch:s=0.5,r=-0.5'` (colorblind-friendly and suitable for grayscale printing) 5. **Gradients**: `'light:<color>'`, `'dark:<color>'` 6. **Blends**: `'blend:<color1>,<color2>'` 7. **Lists of color codes**: hexadecimal codes such as `['#FF5733', '#33FF57']`, RGB tuples, or Matplotlib color names ```{python} # Default palette ax = sns.barplot(data=df, y='island', x='body_mass_g', errorbar=None, hue='sex') ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island') sns.move_legend(ax, bbox_to_anchor=(1, 1), loc='upper left') plt.show() ``` ```{python} # palette='flare': use a predefined Seaborn palette ax = sns.barplot(data=df, y='island', x='body_mass_g', errorbar=None, hue='sex', palette='flare') ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island') sns.move_legend(ax, bbox_to_anchor=(1, 1), loc='upper left') plt.show() ``` ```{python} # Pass a list of Matplotlib color names to palette ax = sns.barplot(data=df, y='island', x='body_mass_g', errorbar=None, hue='sex', palette=['blue', 'orange']) ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island') sns.move_legend(ax, bbox_to_anchor=(1, 1), loc='upper left') plt.show() ``` ```{python} # Pass a list of hexadecimal color codes to palette ax = sns.barplot( data=df, y='island', x='body_mass_g', errorbar=None, hue='sex', palette=['#a1c9f4', '#8de5a1'], ) ax.set(title='Penguin', xlabel='Body Mass (g)', ylabel='Island') sns.move_legend(ax, bbox_to_anchor=(1, 1), loc='upper left') plt.show() ``` **References:**\ [Seaborn `sns.set_theme()` documentation](https://seaborn.pydata.org/generated/seaborn.set_theme.html) (`palette` parameter)\ [Seaborn `sns.color_palette()` documentation](https://seaborn.pydata.org/generated/seaborn.color_palette.html) (list of predefined palette names)\ [Matplotlib named colors](https://matplotlib.org/stable/gallery/color/named_colors.html)\ [Matplotlib colormaps](https://matplotlib.org/stable/users/explain/colors/colormaps.html)\ [w3schools color picker](https://www.w3schools.com/colors/colors_picker.asp) ## Extension: Counting Penguins on Each Island {.unnumbered} The following two approaches are equivalent and can be used later to sort bar plots by totals. **Method 1: `df.groupby().size()`** ```{python} # Group by island and count the number of rows in each group df.groupby('island').size() ``` **Method 2: `df.value_counts()`** ```{python} # Count the occurrences of each value in the island column df.value_counts('island') ``` # Drawing Other Plots ## Point Plot The y-axis of a bar plot must start at 0 to ensure that the visual proportions match the data proportions, following Wilke's [Principle of Proportional Ink](https://clauswilke.com/dataviz/proportional-ink.html) ([Wilke 2019, chap. 17](https://clauswilke.com/dataviz/proportional-ink.html)). Point plots do not have this restriction: the y-axis can start from the data range, and the data-ink ratio is higher because there is less visual clutter. When the x-axis is nominal data, such as island names, the points do not need to be connected with lines. ```{python} # linestyle='none': do not connect points when the x-axis is categorical. # The vertical segments around points are error bars. sns.pointplot(data=df, x='island', y='body_mass_g', hue='sex', linestyle='none') plt.show() ``` **Reference: [`sns.pointplot()` documentation](https://seaborn.pydata.org/generated/seaborn.pointplot.html#)** ## Histogram The x-axis of a bar plot represents **discrete categories** (nominal or ordinal data). The x-axis of a histogram represents **continuous numerical intervals** (ratio data). ```{python} sns.histplot(data=df, x='flipper_length_mm') plt.show() ``` ```{python} # Group with hue: plot penguins from each island separately and overlay the distributions sns.histplot(data=df, x='flipper_length_mm', hue='island') plt.show() ``` ```{python} # Use a consistent y-axis range to compare the plot with and without hue grouping sns.histplot(data=df, x='flipper_length_mm') plt.show() sns.histplot(data=df, x='flipper_length_mm', hue='island') plt.ylim(0, 80) plt.show() ``` ```{python} sns.histplot( data=df, x='flipper_length_mm', hue='island', stat='density', # The total area of all bins is 1 (probability density) common_norm=False # Normalize each group separately. If True, all groups share one total area of 1. ) plt.show() ``` ### The `stat` Parameter This parameter controls the statistic shown by the histogram. Available values include: - `'count'`: bin height = number of data points in the bin. - `'frequency'`: bin height = count / bin width, or frequency per unit width. This is useful when bin widths are unequal. - `'probability'` / `'proportion'`: bin height = count / total count, so all bin heights sum to 1. - `'percent'`: bin height = (count / total count) * 100, so all bin heights sum to 100. - `'density'`: bin height = frequency / bin width, so the total **area** of all bins is 1. As bin width approaches 0, this becomes a smooth density curve. ```{python} # stat='count': bar height = number of data points in each bin ax = sns.histplot(data=df, x='flipper_length_mm', stat='count') # Iterate over each bar (patch), get its height, and show a value label above it for p in ax.patches: # Get the bar height, which is the frequency in the interval height = p.get_height() # Add text ax.text( p.get_x() + p.get_width() / 2, # x coordinate: center of the bar height, # y coordinate: bar height int(height), # text: integer frequency ha='center', va='bottom', # horizontal/vertical alignment ) plt.show() ``` ```{python} # stat='frequency': bar height = count / bin width ax = sns.histplot(data=df, x='flipper_length_mm', stat='frequency') for p in ax.patches: height = p.get_height() ax.text( p.get_x() + p.get_width() / 2, height, int(height), ha='center', va='bottom', ) plt.show() ``` ```{python} # stat='probability': bar height = count / total count, and all bar heights sum to 1 ax = sns.histplot(data=df, x='flipper_length_mm', stat='probability') for p in ax.patches: height = p.get_height() ax.text( p.get_x() + p.get_width() / 2, height, f'{height:.2f}', # Keep two decimal places ha='center', va='bottom', ) plt.show() ``` ```{python} # stat='percent': bar height = (count / total count) * 100, and all bar heights sum to 100 ax = sns.histplot(data=df, x='flipper_length_mm', stat='percent') for p in ax.patches: height = p.get_height() ax.text( p.get_x() + p.get_width() / 2, height, int(height), ha='center', va='bottom', ) plt.show() ``` ```{python} # stat='density': bar height = frequency / bin width, and the total bin area is 1 ax = sns.histplot(data=df, x='flipper_length_mm', stat='density') for p in ax.patches: height = p.get_height() ax.text( p.get_x() + p.get_width() / 2, height, f'{height:.3f}', ha='center', va='bottom', ) plt.show() ``` ### The `common_norm` Parameter This boolean parameter only applies when `stat` is `'probability'`, `'proportion'`, `'percent'`, or `'density'`. Its default value is `True`: - `common_norm=True`: normalize using the **entire dataset**, so bin heights can be compared across groups. - `common_norm=False`: normalize each **subgroup independently**, so bin heights are only meaningful within each group. ```{python} sns.histplot( data=df, x='flipper_length_mm', hue='island', stat='density', common_norm=True, # Overall normalization: the total bin area across all groups is 1 ) plt.show() ``` ```{python} sns.histplot( data=df, x='flipper_length_mm', hue='island', stat='density', common_norm=False, # Separate normalization: each group's bin area sums to 1 ) plt.show() ``` When overlapping groups are difficult to read, use `data=df[df['column']==value]` to filter the data and plot each group separately: ```{python} # Plot by island to avoid overlapping groups and inspect each island's distribution separately ax = sns.histplot( data=df[df['island'] == 'Torgersen'], x='flipper_length_mm', stat='density' ) ax.set(title='Torgersen Island', xlabel='Flipper Length (mm)') plt.show() ax = sns.histplot( data=df[df['island'] == 'Biscoe'], x='flipper_length_mm', stat='density' ) ax.set(title='Biscoe Island', xlabel='Flipper Length (mm)') plt.show() ax = sns.histplot( data=df[df['island'] == 'Dream'], x='flipper_length_mm', stat='density' ) ax.set(title='Dream Island', xlabel='Flipper Length (mm)') plt.show() ``` **Reference: [`sns.histplot()` documentation](https://seaborn.pydata.org/generated/seaborn.histplot.html)** (`hue`, `stat`, and `common_norm` parameters) ## Box Plot A box plot summarizes a distribution with five key values: min, Q1, Q2 (median), Q3, and max. It also marks outliers beyond the whisker range. - **Box**: from Q1 to Q3, containing the middle 50% of the data. - **Line inside the box**: the median (Q2). - **Whiskers**: extend outward from Q1 and Q3, by default to the extreme values within 1.5 * IQR. - **Outliers**: points beyond the whisker range, marked separately. ![Normal distribution and IQR](../../../src/img/ttl/amt-dist/normal-dist.png) **Box Plot vs. Histogram** ![Box plot](../../../src/img/ttl/amt-dist/boxplot.png) Box plots are useful for quickly identifying a distribution summary and outliers. Histograms are better for seeing the frequency distribution within specific intervals. ```{python} sns.boxplot(data=df, x='flipper_length_mm', y='island', hue='sex') plt.show() ``` ```{python} sns.boxplot( data=df, x='flipper_length_mm', y='island', hue='sex', fliersize=8, # fliersize: size of outlier points (default: 5) ) plt.show() ``` ```{python} sns.boxplot( data=df, x='flipper_length_mm', y='island', hue='sex', fliersize=8, whis=2, # whisker: extend to Q1/Q3 +/- 2*IQR (default: 1.5); larger values reduce outlier points ) plt.show() ``` **Reference: [`sns.boxplot()` documentation](https://seaborn.pydata.org/generated/seaborn.boxplot.html)** (`fliersize` and `whis` parameters) ## Violin Plot A violin plot combines features of a box plot, such as medians and quantiles, with a density curve that shows the smoothed distribution. When the dataset is small, the smoothed curve may be misleading, so a histogram or box plot may be preferable. ```{python} sns.violinplot(data=df, y='island', x='flipper_length_mm', hue='sex') plt.show() ``` ```{python} # split=True: combine the two hue groups in the same category into one violin for easier distribution comparison sns.violinplot( data=df, y='island', x='flipper_length_mm', hue='sex', split=True, # Combine two hue groups into one violin for easier comparison ) plt.show() ``` **Reference: [`sns.violinplot()` documentation](https://seaborn.pydata.org/generated/seaborn.violinplot.html)** (`split` parameter) ## Strip Plot and Swarm Plot **Strip plots** draw every data point, with each dot representing one observation. Points are randomly jittered along the categorical axis to reduce overlap. **Swarm plots** automatically arrange overlapping data points and are easier to read. Note that both plots show a univariate **distribution**, meaning how densely one variable appears across its range. This is different from a scatter plot, which shows the **relationship** between two variables. | Comparison | Strip Plot | Swarm Plot | |-------------------|----------------------------|-------------------------| | Basic idea | Randomly jitter data points along the categorical axis to avoid overlap | Precisely arrange points with an algorithm to avoid overlap and show the distribution | | Point placement | Random jitter; overlap may still occur | Non-random; automatically adjusted by density to avoid overlap | | Readability | Clear for small datasets; can overlap heavily with large datasets | Clearer for showing distribution shape in small to medium datasets | | Computational cost | Lower and faster to draw | Higher and slower for large datasets | | Suitable uses | Quick visualization, small datasets, situations where exact distribution shape is less important | Showing distribution details for medium-sized datasets | | Distribution detail | Moderate; true density can be hard to see | Stronger; data density is easier to read visually | | Parameter control | Jitter can be controlled with `jitter` | No jitter needed; the algorithm adjusts placement automatically | | Example function | `sns.stripplot()` | `sns.swarmplot()` | ```{python} sns.stripplot(data=df, x='flipper_length_mm', y='island', hue='sex') plt.show() ``` ```{python} sns.swarmplot(data=df, x='flipper_length_mm', y='island', hue='sex') plt.show() ``` **References:**\ [`sns.stripplot()` documentation](https://seaborn.pydata.org/generated/seaborn.stripplot.html)\ [`sns.swarmplot()` documentation](https://seaborn.pydata.org/generated/seaborn.swarmplot.html) # Reference Links {.unnumbered} > **General links:**\ > [**Python 3 documentation**](https://docs.python.org/3/library/functions.html) (for built-in functions and related usage)\ > [**Pandas documentation**](https://pandas.pydata.org/docs/reference/frame.html) (DataFrame)\ > [**Matplotlib documentation**](https://matplotlib.org/stable/api/pyplot_summary.html) (`plt`)\ > [**Seaborn documentation**](https://seaborn.pydata.org/generated/seaborn.barplot.html) (`barplot`) > > [**`sns.barplot()` documentation**](https://seaborn.pydata.org/generated/seaborn.barplot.html) (drawing bar plots with Seaborn)\ > [**`sns.countplot()` documentation**](https://seaborn.pydata.org/generated/seaborn.countplot.html) (drawing count plots with Seaborn) > > **Changing chart themes:**\ > [**`plt.xticks()` documentation**](https://matplotlib.org/stable/api/_as_gen/matplotlib.pyplot.xticks.html) (the `kwargs` section explains how to rotate tick labels)\ > [**`sns.set()` documentation**](https://seaborn.pydata.org/generated/seaborn.set.html) (equivalent to `sns.set_theme()`)\ > [**`sns.set_theme()` documentation**](https://seaborn.pydata.org/generated/seaborn.set_theme.html#seaborn.set_theme) (set the global Seaborn plotting theme)\ > [**`ax.set()` documentation**](https://matplotlib.org/stable/api/_as_gen/matplotlib.axes.Axes.set.html) (set titles, axis labels, and related elements for a specific Matplotlib axes object) > > **Changing legends:**\ > [**`sns.move_legend()` documentation**](https://seaborn.pydata.org/generated/seaborn.move_legend.html) (Seaborn function for moving legends; see the `obj` and `loc` parameters)\ > [**`matplotlib.axes.Axes.legend()` documentation**](https://matplotlib.org/stable/api/_as_gen/matplotlib.axes.Axes.legend.html) (Matplotlib legend function; its options also apply to `sns.move_legend()`; see `loc`, `ncols`, and `bbox_to_anchor`) > > **Changing chart colors:**\ > [**Seaborn `sns.color_palette()` documentation**](https://seaborn.pydata.org/generated/seaborn.color_palette.html) (predefined palette names for the `palette` parameter)\ > [**Matplotlib named colors**](https://matplotlib.org/stable/gallery/color/named_colors.html)\ > [**Matplotlib colormaps**](https://matplotlib.org/stable/users/explain/colors/colormaps.html)\ > [**Matplotlib colormap guide**](https://matplotlib.org/stable/users/explain/colors/colormaps.html)\ > [**w3schools color picker**](https://www.w3schools.com/colors/colors_picker.asp) (view hexadecimal codes for colors) > > **Other charts:**\ > [**`sns.pointplot()` documentation**](https://seaborn.pydata.org/generated/seaborn.pointplot.html#) (drawing point plots)\ > [**`sns.histplot()` documentation**](https://seaborn.pydata.org/generated/seaborn.histplot.html) (drawing histograms; see `hue`, `stat`, and `common_norm`)\ > [**`sns.boxplot()` documentation**](https://seaborn.pydata.org/generated/seaborn.boxplot.html) (drawing box plots; see `fliersize` and `whis`)\ > [**`sns.violinplot()` documentation**](https://seaborn.pydata.org/generated/seaborn.violinplot.html) (drawing violin plots)\ > [**`sns.stripplot()` documentation**](https://seaborn.pydata.org/generated/seaborn.stripplot.html) (drawing distribution scatter plots)\ > [**`sns.swarmplot()` documentation**](https://seaborn.pydata.org/generated/seaborn.swarmplot.html) (drawing clustered distribution scatter plots) > > **Textbook: *Fundamentals of Data Visualization* (Wilke, 2019):**\ > [**6.1 Bar Plot**](https://clauswilke.com/dataviz/visualizing-amounts.html#bar-plots) (rotated tick labels are harder to read; a horizontal layout is preferable)\ > [**17. The Principle of Proportional Ink**](https://clauswilke.com/dataviz/proportional-ink.html) (the proportions of graphical marks should match the true proportions in the data; for example, bars in a bar plot should start from 0) [⇧ Back to top](#)