Data Wrangling·Lesson 33 of 67·intermediate
Plotting with matplotlib
Draw the four charts that answer most questions about a model — line, bar, histogram and scatter — and read the distribution a summary statistic hides.
In this lesson · 18 min
- →Draw a figure with the explicit fig and ax interface
- →Choose between line, bar, histogram and scatter
- →Label every axis, and know why an unlabelled chart is useless
- →Plot straight from a DataFrame onto an axis you control
- →Put several charts in one figure, and use a log scale when the data demands it
Charts in these lessons render underneath the code: draw a figure and it appears in the output
panel. You do not need plt.show() here — although you do in a script on your own machine, so
the examples include it where it belongs.
The figure and the axes
Always create both explicitly. fig is the canvas; ax is the plot on it, and every method you
want lives on ax:
Lines: something over something else
The gap opening between two curves is the picture of overfitting, and it is far more legible than the two final numbers.
Bars: comparing categories
DataFrame.plot(..., ax=ax) hands pandas an axis you made, so you keep control of the labels and
the figure while pandas does the drawing.
Histograms: the distribution behind the mean
Scatter: does one thing track another?
Several charts in one figure
plt.subplots(rows, columns) returns an array of axes:
Log scales
When values span orders of magnitude, a linear axis shows you one bar and four invisible ones:
Plotting a time series
Try it yourself
Draw both series on ax as lines with the labels train and validation, then add the title Loss, the x label step, the y label loss and a legend.
Draw a histogram of latency on ax with 40 bins, then mark the median with a vertical line. Label the x axis milliseconds.
Check yourself
Your latency has a mean of 380 ms and a median of 210 ms. What does a histogram most likely show?
Check yourself
Why prefer fig, ax = plt.subplots() over calling plt.plot() and plt.title() directly?
Recap
- Create the figure and axes explicitly with
fig, ax = plt.subplots(), and call methods onax. - Line for something over time or steps, bar for categories, histogram for a distribution, scatter for a relationship between two numbers.
- A histogram shows what a mean hides — multiple populations, skew, a long tail.
DataFrame.plot(..., ax=ax)lets pandas draw onto an axis you still control.plt.subplots(rows, cols)gives a grid;set_yscale("log")rescues data spanning orders of magnitude.- Title and label every chart, every time.