Review
Core idea
Plots are inspection tools.
Use them to notice patterns, mistakes, outliers, distributions, and computed curves.
Line plots
Use line plots when order matters:
plt.plot(x, y, marker="o")
x and y should have the same length.
Scatter plots
Use scatter plots for paired measurements:
plt.scatter(x, y)
Prediction-versus-target plots are a common ML inspection tool.
Histograms
Use histograms for distributions:
plt.hist(values, bins=5)
Histograms ignore order and show counts by value range.
Plotting arrays
Use imshow for a two-dimensional array:
State what rows and columns mean.
Labels and saving
Useful labels:
Save a figure with:
plt.savefig("loss.png", dpi=150, bbox_inches="tight")
In a local script, plt.show() displays the figure. Save before showing or
closing it when you also need a file.
When comparing separate imshow figures, use common vmin and vmax limits
so equal colors represent equal values.
Visual debugging
After making a plot, write the observation:
The computed curve is linear, but the expected curve bends upward.
This turns a picture into evidence.