Python for Data Science

What is Data Visualization?

• Data visualization is both a science and an art
• Represent data accurately, without being misleading or wrong...
• ... in an aeasthetically pleasing manner

Aesthetics

• A data visualization maps data values into quantifiable features (aesthetics)
• The choice of aesthetics depends on
• The type of data (quantitative, categorical, ordinal, ...)
• The main message you want to deliver

Aesthetics (Example)

• The position along the x axis represents the duration variable
• The position along the y axis represents the balance variable (logarithmic scale)
• The shape represents the education variable
• The color represents the education variable (redundant coding)

Working With Colors in Data Visualizations

• Three use cases
• Distinguish between groups of observations
• Represent data values
• Highlight specific observations
• The choice of color is different depending on the intended use
• Discrete or continuous?
• Diverging, sequential, or qualitative palette?

Python Data Visualization Packages

Installing and Importing Matplotlib and seaborn

Installing Matplotlib and seaborn with conda (recommended)

% conda install matplotlib seaborn

Installing Matplotlib and seaborn with pip

% pip install matplotlib seaborn

Importing Matplotlib and seaborn (in Python scripts or notebooks)

>>> import seaborn as sns
>>> import matplotlib.pyplot as plt

Example Data Visualization With seaborn and Matplotlib

sns.set_style("ticks") # set seaborn style
sns.set_context("talk") # set seaborn context

fig, ax = plt.subplots(figsize=[8, 5]) # set figure size

plot = sns.scatterplot(
    data=bank_sample,
    x="duration", # variable to map to x axis
    y="balance", # variable to map to y axis
    style="education", # variable to map to shape
    hue="education", # variable to map to color
    alpha=0.5 # transparency
)
plot.set(yscale="log") # use log scale for y axis

# set axis and legend titles
plt.xlabel("Last contact duration (seconds)")
plt.ylabel("Balance ($)")
plt.legend(title="Education")

plt.savefig("dataviz_example.svg") # save to disk

Visualizing Distributions

Histograms

• Histograms can be plotted using the histplot() function (cf. documentation)

fig, ax = plt.subplots(figsize=[7, 5])

sns.histplot(
    data=bank,
    x="age",
    binwidth=5, # 5-year bins
    color="steelblue",
    edgecolor="black")

ax.set( # set plot title and axis labels
    title="Age distribution",
    xlabel="Age (years)",
    ylabel="Frequency"
)

sns.despine(offset=5, trim=True)

plt.savefig("histograms.svg")

Density Plots

• Increasingly popular alternative to histograms
• Plot the underlying distribution of the data using a continuous curve (estimated using a kernel density estimation method)
• Can be plotted using the kdeplot() function (cf. documentation)

fig, ax = plt.subplots(figsize=[7, 5])

sns.kdeplot(
    data=bank,
    x="age",
    color="steelblue"
)

ax.set(
    title="Age distribution",
    xlabel="Age (years)"
)

sns.despine(offset=5)

plt.savefig("dataviz-density.svg")

Comparing Multiple Distributions

• Useful for comparing how a quantitative variable is impacted by another categorical variable
• Can be done using a multitude of plot types
• Histograms (not the best)
• Density plots
• Box plots (boxplot())
• Violin plots (violinplot())
• Usually handled by mapping the categorical variable to a color and/or an axis

Box Plots

• Box plots visualize
• The median (line in the middle of the box)
• The first and third quartiles (the limits of the box)
• The minimum and maximum excluding outliers (wiskers outside of the box)
• Outliers (dots outside of the wiskers)
• Can be plotted using the boxplot() function (cf. documentation)

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

sns.boxplot(
    data=bank,
    x="marital",
    y="age",
    width=.3
)

ax.set(
    title="Age distribution by marital status",
    xlabel="Marital status",
    ylabel="Age (years)"
)

sns.despine()

Violin Plots

• Violin plots can be plotted using the violinplot() function (cf. documentation)
• Similar role to box plots (but more attractive)
• Plotted using kernel density estimators (like density plots)
• Can be misleading if sample size is small, ...

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

sns.violinplot(
    data=bank,
    x="marital",
    y="age",
    width=.5
    # hue="marital"

)

ax.set(
    title="Age distribution by marital status",
    xlabel="Marital status",
    ylabel="Age (years)"
)

sns.despine()

Visualizing Interactions Between Quantitative Variables

• Plotting multiple quantitative variables at once is useful for identifying how the influence each other (e.g., linear relationship, ...)
• Mainly done with scatter plots (scatterplot())
• Color, shape, and size can be used to identify different subsets

Visualizing Amounts With Bar Plots

• Bar plots visualize
• Magnitudes of quantitative values (e.g., totals, counts, ...) ...
• ... for a set of categories of a qualitative variable (e.g., marital statuses, education levels, ...)
• Can be plotted using barplot() (cf. documentation)
• When plotting counts, use countplot() instead (cf. documentation)

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

sns.countplot(
    data=bank,
    x="marital"
  )
ax.set(
  title="Number of clients per marital status",
  xlabel="Marital status",
  ylabel="Count"
)

sns.despine()

Grouped and Stacked Bar Plots

• Grouped or stacked bar plots can be used when you are interested in plotting the quantities for two categorical variables at once
• One categorical variable is mapped to an axis
• The second categorical variable is mapped to color

seaborn Theming

• Figures can be easily styled using seaborn's set_theme(), set_style(), and set_context() functions (cf. documentation)

sns.set_style("white")
...

sns.set_style("whitegrid")
...

sns.set_style("dark")
...

sns.set_style("darkgrid")
...

seaborn Theming

• Figures can be easily styled using seaborn's set_theme(), set_style(), and set_context() functions (cf. documentation)

sns.set_context("paper")
...

sns.set_context("notebook")
...

sns.set_context("talk")
...

sns.set_context("poster")
...

Useful References