Plotting Data using Matplotlib

Written and reviewed by Rohit Singh, Computer Science & Commerce faculty at UniDrill (MCA, Osmania University · 6 years teaching). Based on NCERT Informatics Practices Class XII, Chapter 4.

Which is easier to read: 84 numbers or one picture?

A cultural mela records its sales for three weeks, seven days each. That is 21 numbers in a file. Ask "which day sells most?" and you start scanning. Draw the same numbers as a chart and the answer jumps out: the highest bar is Sunday. Data visualisation means showing data as a graph or chart, to see variation, to show relationships between variables and to communicate results to other people. Traffic symbols, a speedometer and a map in an atlas are everyday examples.

In Python the library for this is Matplotlib, and its drawing module is pyplot.

Here plt is just an alias (a short nickname). You could choose any other name, but plt is the convention and the one exam questions use. Install the library with pip install matplotlib.

Choosing the chart: the question decides

The right chart depends on the question and on the type of data. Learn this table first, because most CUET questions are "which plot for which job".

Anatomy of a plot

A figure is the whole window. Inside it sits the plotting area, with a title, an x label and a y label, ticks (the marks on each axis), optional grid lines and a legend (the key that links each colour to its data). Every pyplot function changes something in the current figure. A good chart always has a title, labelled axes and, when more than one series is drawn, a legend.

A first line chart

Worked example: three days of temperature. Plot the maximum temperature of a city on 25, 26 and 27 December and decorate the chart.

The line rises from 8.5 to a peak of 10.5 on 26/12 and falls to 6.8. Without the labels and the title, the same picture would carry much less meaning.

Pause and predict: What happens if you give plt.plot() only one list, such as plt.plot(temp)?

Check your answer: Matplotlib treats the list as the y values and generates the x values itself as 0, 1, 2 and so on (the positions). It does not fail.

Customising a line: marker, colour, width and style

A marker is a symbol drawn at each data point. Pass it with marker=. Colours come from character codes or names with color=. Width and style come from linewidth= (a number, the default is 1) and linestyle= (a string).

Worked example: turn a specification into code. Plot average height (y, in cm) against average weight (x, in kg) for ages 8 to 16, with a green dashed line of width 2, star markers of size 10, axis labels and a title.

A common slip is swapping the two arguments. The first argument of plot() is always x, so plt.plot(df.height, df.weight) would put height on the horizontal axis.

Beyond the textbook. For the exam: write each property as a keyword: marker=, color=, linestyle=, linewidth=. In reality: plt.plot also accepts a compact format string. The call plt.plot(x, y, 'r*--') draws a red line with star markers and dashes in one go, and '-^' means a solid line with upward triangles. Both forms mean the same thing, and the keyword form is the safe one to use in answers.

Plotting straight from a DataFrame: df.plot()

Series and DataFrame objects have their own plot() method. It is a wrapper around pyplot's plot(), so you can mix it with pyplot decorations such as plt.title() and plt.xlabel(). The chart type is chosen with the keyword kind.

The mela sales file MelaSales.csv has the sales of each day for three weeks, plus the name of the day.

Worked example: three coloured lines. Draw the three weeks as lines in red, blue and brown, with a title and axis labels, and name the days on the x axis.

Step by step: df.plot(kind='line', …) draws one line for each numeric column, and the colour list gives them red, blue and brown in order. The legend appears by default and links each colour to its column name. A line chart needs numbers on the x axis, so the plot uses the row index (0 to 6). plt.xticks(ticks, df.Day) then replaces those numbers with the day names. The first argument of xticks is the positions, the second is the text to show.

Beyond the textbook. For the exam: call plt.show() last. In reality: if you also want the picture saved, call plt.savefig('x.png') before plt.show(). Once the window is shown and closed, the figure is cleared, and saving afterwards often gives an empty file.

Bar chart: comparing categories

A line joins points and suggests a trend. When your goal is comparison between the weeks or between categories, use bars. Bar charts can also put text, such as day names, on the x axis. If you do not tell plot() which column to use for x, it uses the row index 0, 1, 2…, so name the column with x=.

You can also customise the bars: color=['red', 'yellow', 'purple'] gives one colour per week, edgecolor='Green' outlines each bar, and linewidth=2 and linestyle='--' style that outline. To make the bars horizontal, use kind='barh'.

Worked example: read the totals. Which day has the best total sales across the three weeks, and which week was best?

On the bar chart this reads as: the Sunday group of bars is the tallest, and the Week 1 bars add up to the most.

Histogram: how often do values fall in each range?

A histogram looks like a bar chart but each column covers a range of values (a bin), and the column's height is the number of data points in that range. The bars touch each other because the ranges are continuous. df.plot(kind='hist') chooses the bins itself, or you set them with bins=: a number such as bins=20, or a list of edges such as bins=[18,19,20,21,22], or range(18,25).

Worked example: counting into bins. Six heights are 60, 61, 63, 65, 61 and 60 cm. Use the bin edges 60, 62, 64 and 66. How tall is each column?

The heights of the three columns are 4, 1 and 1. Customising is done with keywords: edgecolor='Green' for the border, linewidth=2, linestyle=':', fill=False for empty bars and hatch='o' for a pattern. In df.plot(kind='hist') only numeric columns are drawn, so a text column such as Name is skipped.

Pause and predict: A histogram of 117 yearly minimum temperatures has a tall column in the middle and short columns at both ends. What does it tell you about the temperatures?

Check your answer: Most years had a minimum close to the middle value, and very low or very high minimums were rare. A histogram shows the frequency pattern (the distribution), which a line chart of the same values would not show so quickly.

A frequency polygon is a line drawn through the midpoints of the tops of the histogram bars. You compute the heights and bin edges with np.histogram, find each midpoint as the average of its two edges (mid = 0.5 * (edges[1:] + edges[:-1])) and draw it over the histogram with plt.plot(mid, y, '-^').

Open data portals, such as the Government of India's data.gov.in, publish large datasets that anyone can download and plot. A temperature dataset, for example, can be read with pd.read_csv(..., usecols=[...]), where usecols loads only the columns you name.

Scatter plot: is there a relationship?

A scatter plot puts one dot for every pair of values, one variable on each axis, to show whether two variables are related. That is why it is also called a correlation plot. The size, colour or shape of the dots can carry a third variable.

Worked example: discount against sales. A shop offers discounts of 10, 20, 30, 40 and 50 percent over five weeks and records the sales in rupees.

Reading the chart: as the discount rises, sales rise too (a positive relationship), and the jump from 40 to 50 percent doubles the sales from 50000 to 100000. Note that plt.scatter(x=…, y=…) takes arrays directly. If you use df.plot(kind='scatter') instead, you must name the DataFrame columns with x='column1', y='column2', because a DataFrame has many columns.

Box plot: the five number summary in one picture

Quartiles divide sorted data into four equal parts. The first quartile Q1 (25 percent point), the second quartile Q2 (the median, 50 percent) and the third quartile Q3 (75 percent) mark the splits. A box plot draws the summary: the box runs from Q1 to Q3 with a line at the median, the whiskers reach out to the lowest and highest ordinary values, and points beyond the whiskers are shown separately as outliers, values far from the rest.

Worked example: read the box plot of 13 students in five subjects. The marks, summarised, are:

Which subject has the least variation, and which shows outliers? Variation is the size of the box, $\text{IQR} = Q_3 - Q_1$. English has 19, Maths 20, Hindi 16, Science 24 and Social Studies 9. Social Studies has the smallest box, so it is the most consistent subject.

To find outliers, matplotlib uses the rule that a point is an outlier if it lies more than $1.5 \times \text{IQR}$ beyond the box. For Social Studies:

So Social Studies shows two outliers, 54 and 95. Applying the same rule to Maths (fences 13 and 93) gives one outlier, 95. The other subjects have none.

Beyond the textbook. For the exam: an outlier is a value numerically distant from the rest, and the whiskers reach the highest and lowest values. In reality: in matplotlib's default box plot the whiskers stop at the furthest data point within 1.5 times the IQR of the box. Anything beyond is drawn as an outlier dot, which is why "whisker reaches the highest value" is true only when no outlier exists.

For a horizontal box plot, add vert=False, and color='red' changes the colour: df.plot(kind='box', title='Compare Resorts', color='red', vert=False).

Pie chart: parts of a whole

A pie divides a circle into sectors, each sector showing a part of the total. For df.plot(kind='pie') you must choose the column with y= (or set subplots=True, which draws one pie for each numeric column). The label of each sector is, by default, the row index of the DataFrame.

The sectors are labelled Mercury, Venus and Earth because they are the index values. Two keywords customise a pie. explode takes a list of fractions of the radius by which each sector is pulled out, and autopct formats the percentage printed on each sector. legend=False removes the legend, and colors sets the sector colours.

Worked example: forest cover of north eastern states. Draw a pie of forest cover with the first sector exploded by 0.1 and the fifth by 0.2, and percentages to two decimals.

With legend=False the legend box is not drawn. The percentage of each sector is its share of the total forest cover of 170423 sq km.

As a check, take Arunachal Pradesh: $\dfrac{67353}{170423} \times 100 = 39.52$. The explode list has one number for each sector, in the same order as the rows, so the first value 0.1 pulls out Arunachal Pradesh and the fifth value 0.2 pulls out Mizoram.

Pause and predict: You run df.plot(kind='pie') on a DataFrame with two numeric columns and give no y. What happens?

Check your answer: You get an error, because Pandas does not know which column to turn into sectors. Give y='column' or subplots=True.

Exam traps

60-second revision

Always add title, xlabel, ylabel (and legend for several series), then show(). Use savefig('name.png') to save.

Where this chapter connects