Bar Plot

I’d seen this material before but transcribed it here as review.

Principle of Proportion Ink

The actual value and the ink used to represent it should be proportional.

• A principle that applies across all visualizations.
• The x-axis must always start at 0.
  • The vertical ratio of the plot should represent the data differences.


The left chart doesn’t start from 0. So the graph’s proportions fail to reflect the actual data proportions. To avoid unnecessary misunderstanding, use a chart like the one on the right.

Bar Plot

• Data represented with bars, as the name suggests.
• Well suited for comparing numerical values across categories.

Classification by Bar Direction

In matplotlib:

• .bar() : categories on x-axis, data on y-axis
• .barh() : categories on y-axis, values on x-axis
  • Better when there are many categories

Multiple Categories

• Bar plots represent values for a single feature per category
  • Only shows values for 1 feature
  • To display multiple groups, other methods are needed

Solutions:

• Multiple plots
• Represent simultaneously in a single plot
  • Stacked
  • Overlapped
  • Side by side

Stacked Bar Plot


• Stack 2 or more groups
• Distribution of the bottom-most stacked bar is easy to see.
  • Bars stacked above it are harder to interpret.
• matplotlib
  • .bar() uses the bottom parameter for stacking.
  • .barh() uses the left parameter for stacking.

Percentage Stacked Bar Chart


Since stacked bar plots make the overall distribution hard to see, this converts the stacked bar chart to percentages.

Overlapped Bar Plot


• Good for comparing just 2 groups.
  • Compare by adjusting transparency (alpha).
  • Keep in mind that color brightness and saturation also affect the transparency effect.
• Not great for 3 or more groups.
  • Area plots work better in that case.

Grouped Bar Plot

The most effective method! 

• Place group bars side by side
• Implementation is tricky in matplotlib, so handled in seaborn.
  • Uses .set_xticks(), .set_xticklabels().

All the methods above work well with 5-7 groups. If there are more groups, different approaches are needed.

Data Sorting

• Sorting is a must
  • e.g., in pandas: sort_values(), sort_index()
• Sort appropriately by time, size, category order, or category value.

Proper Use of Space

• matplotlib’s bar plot fills the ax and feels cramped
• Adjust with the following methods:
  • X/Y axis limit (.set_xlim(), .set_ylim())
  • Spines (.spines[spine].set_visible())
  • Gap (width)
  • Legend (.legend()), where to place it
  • Margins (.margins())