Python Charts

Python plotting and visualization demystified

Pandas

Plot Stacked vs Unstacked Bar Charts in Pandas

Plot stacked and unstacked bar charts in Pandas with df.plot and the stacked argument.

A bar chart with multiple series can be drawn side by side (unstacked, or grouped) or piled on top of each other (stacked). Pandas switches between the two with a single stacked argument on df.plot(kind="bar").

Quick Example

Grouped bars are the default; pass stacked=True to stack them.

import pandas as pd

df.plot(kind="bar")                  # unstacked (grouped)
df.plot(kind="bar", stacked=True)    # stacked

Both need the same wide DataFrame: one column per series, with the category on the index.

Set up wide data

Bar charts need one column per series. If your data is long (one row per quarter and channel), widen it first so each channel becomes a column.

wide = long.pivot(index="quarter", columns="channel", values="spend")

Each channel is now a column and each quarter is a row, which is exactly the shape df.plot(kind="bar") expects.

Unstacked (grouped) bars

The default, with stacked absent or False, draws one thin bar per series per category, grouped next to each other.

wide.plot(kind="bar", color=["#6d597a", "#b56576", "#e56b6f", "#eaac8b"], rot=0)

Unstacked grouped bar chart of marketing spend by channel per quarter

Grouped bars make it easy to compare series against each other within a single category, but they can get crowded when there are many series.

Stacked bars

Pass stacked=True and each category's series pile into a single bar whose total height is the sum.

wide.plot(kind="bar", stacked=True, color=palette, rot=0)

Stacked bar chart of marketing spend by channel per quarter

Stacked bars are ideal for showing a total per category alongside how it is split by series. The downside is that comparing the size of individual slices across categories is harder, because the slices do not share a common baseline.

Reshape with unstack()

If your data already sits in a MultiIndex, the unstack() method moves one index level into the columns, producing the wide form Pandas needs.

wide = long.set_index(["quarter", "channel"]).unstack("channel")["spend"]

Taken together, pivot and unstack are the two main ways to go from long to wide before plotting. Use whichever reads more naturally for your data.

Make a 100% stacked bar

To compare the share each series contributes rather than raw totals, normalize every category to sum to 100 first.

pct = wide.div(wide.sum(axis=1), axis=0) * 100
pct.plot(kind="bar", stacked=True, rot=0)

100 percent stacked bar chart showing the share of spend per quarter

Now each bar is the same total height, and the segment heights show the percentage split per quarter. This is the chart for "how does the composition differ" instead of "how big is the total".

Try it horizontally

Swap kind="bar" for kind="barh" to draw the bars horizontally. Stacking works the same way.

wide.plot(kind="barh", stacked=True, color=palette)

Horizontal stacked bar chart of spend by channel per quarter

Horizontal stacked bars are handy when the category labels are long and would be cramped along the bottom.

Practical Tips

  • Use grouped (unstacked) bars to compare series within each category.
  • Use stacked bars to show totals and their breakdown.
  • unstack() or pivot to reshape long data into the wide form before plotting.
  • Normalize rows with div(..., axis=0) to build a 100% stacked chart.
  • Swap to kind="barh" when labels are long.

Similar Topics