Debugging Visualizations

A blank chart, a jumbled line, an empty legend — these are the everyday frustrations of visualization. The good news: most visualization bugs fall into a small number of recognizable patterns. Learn the patterns and you'll stop being stuck.

The mindset: "the chart is honest about the data"

If your chart looks wrong, the chart is probably showing the data correctly — but the data isn't what you thought it was. The first move is rarely "fix the chart code." It's usually inspect the DataFrame.

A debugging session almost always starts like this:

Those four lines of inspection solve half of all visualization bugs.

Common failure modes

1. Wrong dtype on the axis

You expected a line chart of dates, but your "date" column is a string. Plotly will treat each value as a category and order them alphabetically. Fix:

2. Unsorted x in a line chart

Lines connect points in the order the DataFrame gives them. If the data isn't sorted by x, you get a "scribble." Always sort_values() before a line chart.

3. Long-vs-wide format confusion

Plotly Express expects long format: one row per observation, columns for category/value/time. If you have a wide table (columns are months, say), reshape with df.melt(...) first.

4. Empty chart after filtering

You filter the DataFrame and Plotly renders an empty canvas. Two likely causes:

• Your filter condition matches zero rows (typo? wrong comparison?).
• You filtered on a column that has NaN values, which silently drop out of comparisons.

Check len(df_filtered) before plotting.

5. Missing values warping the chart

A single NaN in a line chart creates a gap. A single huge outlier compresses every other point against the floor. Two fixes:

• df.dropna(subset=["y"]) — drop missing rows for the columns you're plotting.
• Inspect the y-range with df["y"].describe() to spot outliers.

6. Log of zero or negative

px.scatter(..., log_y=True) will silently drop zero and negative values (you can't take log of them). If your chart suddenly has fewer points than your DataFrame, this is a likely culprit.

7. Too many categories crowding the legend

If color="city" has 200 cities, the legend becomes useless and the colors become indistinguishable. Two fixes:

• Aggregate / group small categories into "Other."
• Use faceting instead of color when categories are many.

8. Mismatched data and figure types

Trying to make a line chart from a categorical x-axis usually indicates you meant a bar chart. Trying to make a scatter from two categorical columns usually indicates you meant a heatmap or grouped bar.

The debugging loop

When a chart looks wrong:

1. Inspect the DataFrame. head(), dtypes, shape, isna().sum(), describe().
2. Check the column types. Numeric where you expect numeric? Datetime where you expect datetime?
3. Sort the data if you're plotting a line.
4. Filter and re-inspect before plotting.
5. Simplify the chart — drop color, facet, size and see if the basic shape is right.
6. Add encodings back one at a time.

This loop will resolve nearly every "my chart looks wrong" moment you'll encounter as a beginner.

A worked example

Check your understanding

QuestionSelect one

Your line chart looks like a scribble — lines crossing back and forth instead of going left-to-right. The most likely cause is:

A bug in Plotly.

The wrong template.

The DataFrame is not sorted by the x-axis column. Lines connect points in DataFrame order, not in x-axis order.

You need more colors.

QuestionSelect one

Plotly Express generally expects data in:

Wide format (one column per category).

Long format (one row per observation, with separate columns for the category, the value, and any grouping variable).

A NumPy array.

A nested dictionary.

QuestionSelect one

You add log_y=True to a scatter plot and suddenly some points disappear. Why?

Plotly randomly samples points on log axes.

Log of zero or a negative number is undefined, so those points are silently dropped.

Log axes only show 100 points.

The simple_white template hides points.

QuestionSelect one

When a chart looks wrong, the first debugging step should be:

Try a different chart type.

Try a different template.

Inspect the DataFrame: head(), dtypes, shape, isna().sum(). The chart is usually showing the data correctly; the data isn't what you think it is.

Restart the notebook.