Line Graphs and Time Series

Line Graphs and Time Series

Contents
1. Introduction.........................................................................................................1
2. Graphs.................................................................................................................2
3. Line Graphs.........................................................................................................6
4. Matplotlib..........................................................................................................10
5. Customizing a Graph........................................................................................13
6. WHO Time Series Data....................................................................................16
7. Types of Growth...............................................................................................18
8. Types of Change...............................................................................................22
9. Comparing Line Graphs....................................................................................24

1. Introduction
At the heart of any data science workflow is data exploration. Most commonly, we
explore data by using the following:

 Statistical methods (measuring averages, measuring variability, etc.)

 Data visualization (transforming data into a visual form)

This indicates that one of the central tasks of data visualization is to help us
explore data.

The other central task is to help us communicate and explain the results we've
found through exploring data. That being said, we have two kinds of data
visualization:

 Exploratory data visualization: we build graphs for ourselves to explore data

and find patterns.
 Explanatory data visualization: we build graphs for others to communicate
and explain the patterns we've found through exploring data.

© Aptech Ltd. 1
 Line Graphs and Time Series

We're going to learn the following:

 How to visualize time series data with line plots.

 What are correlations and how to visualize them with scatter plots.
 How to visualize frequency distributions with bar plots and histograms.
 How to speed up our exploratory data visualization workflow with the
pandas library.
 How to visualize multiple variables using Seaborn's relational plots

2. Graphs
Before we get into Matplotlib and start exploring a dataset, we'll go through a brief
introduction to graphs — what they are and how to build them mathematically.

We can create a graph by drawing two lines at right angles to each other. Each line
is called an axis — the horizontal line at the bottom is the x-axis, and the vertical
line on the left is the y-axis. The point where the two lines intersect is called
the origin.

© Aptech Ltd. 2
 Line Graphs and Time Series

Each axis has length — below, we see both axes marked with numbers, which
represent unit lengths.

The length of the axes helps us precisely locate any point drawn on the graph.
Point A on the graph below, for instance, is seven length units away from the y-axis
and two units away from the x-axis

© Aptech Ltd. 3
 Line Graphs and Time Series

The two numbers that represent the distances of a point from the x- and y-axis are
called coordinates. Point A above has two coordinates: seven and two. Seven is
the x-coordinate, and two is the y-coordinate.
The coordinates often appear in the form (x, y), with the x-coordinate first. So the
coordinates of A are (7, 2). So, here's what we need to know about coordinates:

 The x-coordinate shows the distance in unit lengths relative to the y-axis.
 The y-coordinate shows the distance in unit lengths relative to the x-axis.

The unit lengths of the x- and y-axes doesn't have to be the same. Below, we see
the unit of length on the x-axis is 10, while on the y-axis it is 1,000 (note that we can
also hide some of the numbers to make the graph look better).

© Aptech Ltd. 4
 Line Graphs and Time Series

Examine the graph below, and answer the following questions:

© Aptech Ltd. 5
 Line Graphs and Time Series

3. Line Graphs

Previously, we went through a quick introduction to graphs. On this screen, we're

going to create a graph using a small dataset.
Below, we see a table showing the number of new COVID-19 infections reported
world-wide for the first seven months of 2020:

© Aptech Ltd. 6
 Line Graphs and Time Series

Source: https://covid19.who.int/

Each row shows a pair of two connected data points

 The month number (where one means January, two means February, and so
on)
 The number of cases reported for that month

When we have a pair of two numbers, we can map it on a graph by using the two
numbers as coordinates. Below, we added a point corresponding to the
coordinates (5, 2835147) — this corresponds to the month of May. Behind the
curtains, we generated the graph using Matplotlib, which we'll introduce on the
next screen

© Aptech Ltd. 7
 Line Graphs and Time Series

Let's now put all the data in the table on the graph following the same method:

When we graph how something changes over time, we connect all the points with a
line — above, we graphed how the number of new COVID-19 cases changed month
by month.

© Aptech Ltd. 8
 Line Graphs and Time Series
Because we use lines to connect the points, the graph above is a line graph (also
known as a line plot, or line chart; the distinction between graph, plot and chart is
ambiguous in the data visualization world, with different authors assigning slightly
different meanings to these terms — in this course, we use all three synonymously).

When we create line graphs, it's common to hide the points on the line:

By looking at the line graph we built for our table above, we can see a few patterns.

Overall, the line shows an upward direction, which indicates the number of new
reported cases has gone up month by month and has never decreased or
stabilized. This is mostly a result of the virus spreading. Countries also started to
test more people, which increased the number of new reported cases.

The line connecting January to March has a moderate upward steepness (the
January-February line is almost horizontal), which indicates a moderate increase in
the number of new reported cases. In that period, the virus was just starting to
spread around the world, and many countries were testing people only when they
got to the hospital.

The March-April line is very steep, indicating a surge in new reported cases. The
April-May line shows a mild steepness, so the number of new cases remained high
(around three million). However, the number didn't increase too much compared
to April — this is most likely due to the worldwide lockdowns

The May-July line is very steep, indicating another surge in the number of cases
(from about three million to approximately seven million). This is most likely
because of ending the lockdowns, which created the conditions for more virus
spreading.

Learning how to interpret graphs is just as important as knowing how to build

them. In the exercise below, we'll look at another line graph and interpret it. On the
next screen, we'll learn how to build a line graph using Matplotlib

Below, we see a line graph showing how the number of new reported deaths has
evolved by month in the January-July interval. Examine the graph and then
evaluate the truth value of the following sentences:

© Aptech Ltd. 9
 Line Graphs and Time Series

4. Matplotlib
On the previous screen, we learned about line graphs, but we haven't yet discussed
how to create one with code. Recall that we examined a line graph showing the
evolution of new reported cases over the first seven months of 2020.

© Aptech Ltd. 10
 Line Graphs and Time Series

We can build this line graph ourselves using Matplotlib, a Python library specifically
designed for creating visualizations. Let's start by importing Matplotlib.

A quirk of Matplotlib is that we generally import the pyplot submodule instead of the
whole module:import matplotlib.pyplot instead of import matplotlib.

When we import matplotlib.pyplot, we need to use the plt alias, by convention (import
matplotlib.pyplot as plt).

The pyplot submodule is a collection of high-level functions we can use to generate

graphs very quickly. To create our line graph above, we need to:

 Add the data to the plt.plot() function.

 Display the plot using the plt.show() function.

© Aptech Ltd. 11
 Line Graphs and Time Series

We see a rather odd "1e6" sign on the top left section of the graph. This is scientific
notation, and it tells us that the values on the y-axis are multiplied by 10 6. This
means that a seven on the y-axis means 7 multiplied by 10 6, which is seven million
— we'll get back to this on the next screen.

The plt.plot() function generates a line graph by default. All it needs is two arrays of
data of the same length — these can be Python lists, pandas Series, NumPy arrays,
etc. Above, we used two Python lists.

Notice the order of arguments in plt.plot(month_number, new_cases): month_number comes

first, followed by new_cases. The array that comes first gives the x-coordinates, and
the second array gives the y-coordinates.

The two arrays must be equal in length, or some coordinates will remain unpaired,
© Aptech Ltd. 12
 Line Graphs and Time Series
and Matplotlib will raise an error.

Let's create a new line graph in the exercise below. On the next screen, we'll learn
more about customizing the graph: adding a title, axes labels, and removing the
"1e6" notation.

5. Customizing a Graph

On the previous screen, we built a line graph showing the evolution of new cases
by month:

© Aptech Ltd. 13
 Line Graphs and Time Series

On the top left side of the graph, we see an "1e6" sign — this is scientific notation.
Matplotlib changes to scientific notation if one value on the axis needs to be one
million or greater. If we want to remove scientific notation, we can use
the plt.ticklabel_format(axis, style) function.

The axis parameter defines which axis to configure — its arguments are the
strings 'x', 'y', and 'both'.

The style parameter controls the notation style (plain or scientific). Its arguments
are 'sci', 'scientific', and 'plain'.

The next thing we're going to do is use the plt.title() function to add a title to our line
graph.

© Aptech Ltd. 14
 Line Graphs and Time Series

The x-axis shows the month number, and the y-axis shows the number of new
reported cases. We can show this on our graph by adding a label to each axis — a y-
label and an x-label. To add axis labels, we use plt.xlabel() and plt.ylabel()

Adding a title and axis labels is always a good thing — even if we're just exploring
data for ourselves and no one else will ever see our work.

We create many graphs when we explore data, and we often lose track of what
each graph describes. If we plot a graph now and then examine it again forty
minutes later, the title and the axis labels will help us immediately determine what
that graph is about.

Let's customize a graph in the next exercise.

© Aptech Ltd. 15
 Line Graphs and Time Series

6. WHO Time Series Data

On the previous screen, we stored our data in a few Python lists and used them to
generate line graphs. Next, we're going to use a larger dataset that we've collected
from the World Health Organization.

Let's read in the dataset using the pandas library:

© Aptech Ltd. 16
 Line Graphs and Time Series

The dataset contains data from January 4 until July 31. Each row describes the
COVID-19 report for one day in one specific country (the first few rows show only
China because the virus was only present in China at that time)

The rows in our dataset are listed in time order, starting with January 4 and ending
with July 31. We call a series of data points that is listed in time order a time
series.

Typically we visualize time series with line graphs. The time values are always
plotted, by convention, on the x-axis.

Let's read in our dataset in the following exercise.

© Aptech Ltd. 17
 Line Graphs and Time Series

7. Types of Growth
On the previous screen, we read in our dataset and stored it in a variable
named who_time_series. Let's quickly remind ourselves about the dataset's structure:

Italy was the second epicenter of the pandemic after China. Let's see how the total
number of cumulative cases (recall this is different from the number of new cases)
evolved over the first seven months of 2020. In the code below, we begin by
isolating the data for Italy, and then we create the plot.

© Aptech Ltd. 18
 Line Graphs and Time Series

Until March, the number of cumulative cases stays very low. But then the number
starts to grow very fast (the line on the graph goes upwards very rapidly in March),
and it maintains that fast pace until May. The growth then starts to settle down,
and on the graph, we see an almost horizontal line.

Generally, a quantity that increases very quickly in the beginning — and then it
slows down more and more over time — has a logarithmic growth.

In the March-July period (thus excluding January and February), Italy had a
logarithmic growth in the number of cumulative cases because there were many
new cases in the March-April period, but then the number of new cases started to
decrease. The line on the graph will become perfectly horizontal when there will be
no more new cases.

If we look at India, we can see another type of growth:

© Aptech Ltd. 19
 Line Graphs and Time Series

The number of cumulative cases increases very slowly in the February-May period
(the line is almost horizontal). But then the growth becomes fast (the line rapidly
switches direction upwards), and it gets faster and faster over time, without
showing any sign of slowing down.

Generally, a quantity that increases slowly in the beginning — but then starts
growing faster and faster over time — has exponential growth.

India shows exponential growth for the data we have, but when the number of new
cases will decrease, the growth (of cumulative cases) will become logarithmic.

If we look at Italy again, we can actually see an exponential growth too if we isolate
only the February-May period. Overall, Italy has a slow growth in the beginning,
followed by a fast growth in the March-May period, and then the growth slows
down again. This sequence of growth rates is often described as logistic growth.

Now, let's plot a line graph for Poland to see another type of growth:

© Aptech Ltd. 20
 Line Graphs and Time Series

If we look at the April-July period, we can see an approximately straight line. There
are a few variations here and there, but no obvious curves like we see for Italy or
India. The number of cases increases nonetheless, but it increases at a constant
rate.

Generally, a quantity that increases constantly over time has linear growth.

To sum up, these are the three types of growth we've learned in this screen:

We will continue the discussion about types of growth on the next screen. Let's
now look at an exercise.

© Aptech Ltd. 21
 Line Graphs and Time Series

8. Types of Change
On the previous screen, we learned three common types of growth: linear,
exponential, and logarithmic. As a word of caution, labeling a type of growth just
by looking at a graph is far from being precise. These types of growth are best
described by precise and well-defined mathematical functions. However, these
visual approximations can serve as useful mind tools that we can use to interpret
how time series data change.

© Aptech Ltd. 22
 Line Graphs and Time Series

Change is not only about growth. A quantity can also decrease following a linear,
exponential, or logarithmic pattern.

The data, however, rarely fits any of these patterns perfectly. Most often, our line
graphs are only approximately linear, approximately exponential, or approximately
logarithmic. Moreover, one portion of a single line graph can show an exponential
change, another portion of the same graph can show a linear change, while
another can show an irregular change that doesn't resemble any common pattern.

In practice, most of the line graphs we plot don't show any clear pattern. We need
to pay close attention to what we see and try to extract meaning without forcing
the data into some patterns we already know.

If we look at the evolution of new cases in Belarus, for instance, we see many
irregularities on the line graph:

© Aptech Ltd. 23
 Line Graphs and Time Series

In the April-July period, we see several spikes on the graph going either upward or
downward. For some days, the number of new cases gets close to 2,000 (the
upward spikes), while for others is zero (the downwards spikes). These large
variations suggest that the reports didn't arrive daily — it may be that no one sent
reports over the weekends or on national holidays. The number of new cases keeps
increasing until the next report, and then we see one of those upward spikes

When we see irregularities on a line graph, this doesn't mean we can't extract any
meaning. By analyzing the irregularities, we can sometimes uncover interesting
details.

9. Comparing Line Graphs

So far, we've learned what line graphs are, how to build one using Matplotlib, and
some common types of change. Next, we're going to focus on comparing line
graphs.

One of the key elements of data exploration is comparison — how does this value
compare to that other value? For our COVID-19 time series, we can formulate many
questions in terms of comparison:

© Aptech Ltd. 24
 Line Graphs and Time Series

 How does the United Kingdom compare to France with respect to the
evolution of cumulative new cases?
 How does Mexico compare to the United States with respect to the
cummulative number of deaths?
 How does the evolution of new reported cases compare between India,
Indonesia, and China?
 How does the evolution of total cases compare between Europe and Asia? Or
between Africa and South America?
For instance, let's visualize the evolution of cumulative cases for France and the
United Kingdom. Matplotlib allows us to have two line graphs sharing the same x-
and y-axis:

We see two lines of different colors above, but we can't tell which is for France and
which is for the United Kingdom. To solve this problem, we're going to add
a legend that shows which color corresponds to which country. In the code below,
we first add a label argument to the plt.plot() function, and then we use
the plt.legend() function:

© Aptech Ltd. 25
 Line Graphs and Time Series

When we use plt.plot() the first time, Matplotlib creates a line graph. When we
use plt.plot() again, Matplotlib creates another line graph that shares the same x- and
y-axis as the first graph. If we want Matplotlib to draw the second line graph
separately, we need to close the first graph with the plt.show() function

© Aptech Ltd. 26
 Line Graphs and Time Series

Looking at the two graphs above, the evolution of cumulative cases looks very
similar if we only judge by the shape of the line. If we look on the y-axis, however,
we see that the two graphs have different ranges, and the values for the UK are
almost twice as large. It's much easier to compare these two visualizations if they
share the same axes.

Let's now do an exercise and wrap up this lesson on the next screen.

© Aptech Ltd. 27
 Line Graphs and Time Series

In this lesson, we went through a quick introduction to graphs, and then we

learned how to do the following:

 Plot and customize a line graph using Matplotlib

 Visualize time series with line graphs
 Interpret line plots by identifying types of change

In the next lesson, we're going to learn about seasonality, correlation, and scatter
plots

© Aptech Ltd. 28