Make Figures Comparable and Reusable

A figure carries numerical evidence beyond the program that created it. A reader should be able to tell what was measured, which units were used, and whether two visible distances mean the same numerical change. The program must also save the intended figure and release it when its work is done.

We will compare three stages of the recurring measurement calculation:

All three tables have axes (observations, sensors). Their values are raw readings, calibrated readings, and differences from each sensor's calibrated mean. We will measure each quantity in degrees Celsius. The numerical context belongs in the text and labels; a plot should not make the reader recover it from the source code.

Label the Quantities Being Compared

Create the figure and its two axes explicitly:

fig owns the complete figure. axes[0] and axes[1] own their individual panels. Calling methods on those objects makes the target of each operation visible:

“Reading” names the vertical quantity, while “°C” supplies its unit. “Observation” explains the horizontal positions. Labels such as “value” and “index” would leave too much unstated.

A legend is useful when one panel contains more than one series and their identity is not already clear from direct labels. Here each panel contains two sensors, so a legend has work to do:

Markers and line styles distinguish the sensors in addition to color. The comparison remains readable in grayscale and for readers who do not perceive the colors as distinct. A legend for a single line named completely by the title and axis labels would add little.

Keep Numerical Scales Comparable

If raw and calibrated values are placed in separate panels, unrelated vertical limits can make equal numerical changes look different. sharey=True gives both panels the same vertical scale. We can also set the intended limits once:

The shared range means the same vertical distance represents the same number of degrees in both panels. It does not prove that the calibration is correct; the numerical calculation and its checks supply that evidence.

Other plot forms need the same fairness in their own terms:

• histogram comparisons need the same bin edges, not merely the same number of bins;
• heatmap comparisons need the same vmin and vmax, and the same color meaning;
• bar and line comparisons need compatible axis limits and scales;
• every comparison needs matching quantities and units before shared limits can be meaningful.

The following executable example puts all three stages together. Its printed limits are numerical evidence that the panels share a scale, while the panel labels distinguish readings from differences around a mean.

Shared limits support a comparison of numerical magnitudes; they do not turn the three quantities into one quantity. Zero in a centered panel means the sensor's calibrated mean, while zero in a reading panel means a zero reading. The titles and axis labels preserve that distinction.

Save the Intended Figure, Then Close It

An output path should be explicit. Create its directory, save through the known figure object, then close that same object:

fig.savefig(...) cannot accidentally select another active figure. The path states where the file belongs, and the .png extension selects the file format. plt.close(fig) releases this figure from pyplot after the export is complete. Saving must come first.

In Browser Python, the path belongs to the temporary browser-side file system for that page session. In a local program, it is resolved from the program's working directory unless an absolute path is used. Keep the plotting code, source data, transformation, limits, and output path together so the image can be recreated rather than treated as unexplained evidence.

References

• Matplotlib documentation: pyplot.subplots — figure and axes creation, including shared axes.
• Matplotlib documentation: Figure.savefig — saving a specific figure to a path.
• Matplotlib documentation: pyplot.close — closing a specific figure.