More plotting tricks using Altair

cereal = pd.read_csv('data/cereal.csv')

import altair as alt

chart0 = alt.Chart(cereal, width=500, height=300).mark_circle().encode(
    x='mfr', 
    y='calories'
).properties(title="Scatter plot of manufacturer calorie content")

chart0

Let’s build on the Altair skills we learned in the previous module.

At this point, we are familiar with writing basic plotting code similar to what is shown here.

However, it’s important that we start specifying what kind of variable type we use for our x and y values with the encode(..) verb.

Before, Altair would guess what type of data it was plotting. Usually it’s pretty smart and guesses correctly like we saw in our previous plots, but unfortunately this is not always the case.

chart1 = alt.Chart(cereal_modified, width=500, height=300).mark_circle().encode(
                   x='mfr', 
                   y='calories'
         ).properties(title="Scatter plot of manufacturer calorie content")

chart1

Let’s see an example where Altair fails to determine the correct data type.

For this example, we have modified the calories column in the cereal dataframe.

We will now generate a scatter plot of mfr and calories from this modified cereal dataset.

Notice how 150 comes before 100 on the y-axis? It seems we have a problem here, which is due to Altair failing to recognize that calories is a numerical type.

Even Altair can’t always get it right every time, which is why it’s so important we specify the data type when plotting.

chart2 = alt.Chart(cereal_modified, width=500, height=300).mark_circle().encode(
                   x='mfr:N', 
                   y='calories:Q'
                  ).properties(title="Scatter plot of manufacturer calorie content")

chart2

We can help Altair by giving it clear instructions on what type of columns our x and y values are.

In this case we are going to specify N for the nominal column mfr and Q for the quantitative column calories.

That’s better!

Data Type	Shorthand Code	Description	Examples
Ordinal	O	a discrete ordered quantity	“dislike”, “neutral”, “like”
Nominal	N	a discrete un-ordered quantity	eye color, postal code, university
Quantitative	Q	a continuous quantity	5, 5.0, 5.011
Temporal	T	a time or date value	date (August 13 2020), time (12:00 pm)

Altair recognizes the following column types and it’s best practice that we specify this when we plot going forward.

Ordinal values imply that there is some natural ordering to the values.

For example, the ratings of a a movie could be on an ordinal scale since a five star rating is better than a single star rating.

In contrast, there is no such natural ordering for nominal values. An example of this would be someone’s eye colour, their country location or the university they attended.

Anything numeric is considered a quantitative variable and time or date values are considered as temporal.

chart3 = alt.Chart(cereal, width=500, height=300).mark_circle().encode(
                   x='sugars:Q',  # set the sugars column as quantitative
                   y='rating:Q'   # set the rating column as quantitative
         ).properties(title="Scatter plot of cereal rating vs sugar content")

chart3

Let’s practice this.

Maybe we are interested in plotting the rating of cereals vs the amount of sugar they contain from cereal dataframe.

We do this using a scatter plot which uses .mark_circle(). We can assign sugars as the the x variable and ratings as the y variable from the cereal dataframe we have been using.

Here, sugars and rating are both quantitative columns so we specify Q as variable types in our plot.

Variable types

chart4 = alt.Chart(cereal, width=500, height=300).mark_circle().encode(
                   x=alt.X('sugars:Q'), # use alt.X() to map the x-axis
                   y=alt.Y('rating:Q')  # use alt.Y() to map the y-axis
         ).properties(title="Scatter plot of cereal rating vs sugar content")

chart4

So far when plotting with Altair, we have been mapping our x and y in the encode(x=..,y=..) verb.

However, doing so gives us very little control over how exactly we would like to map our x and y values.

In order to have more control, we can map our x and y values using x=alt.X(...) and y=alt.Y(...) respectively.

This gives us a lot more control over the customization of our plot.

You’ll see this coming up.

Histograms

chart5 = alt.Chart(cereal, width=500, height=300).mark_bar().encode(
                   x=alt.X('calories:Q', bin=True), # set x-axis as calories 
                   y=alt.Y('count():Q')             # set the y-axis as the occurrence count for each calorie value
         ).properties(title="Histogram plot of cereal calorie content")
chart5

Another type of plot we can make using Altair is called a histogram.

A histogram would be an ideal plot if we were interested in seeing how many cereals in our dataframe have calories within a certain range. A histogram is a bar chart where the height of each bar shows the frequency of something occurring. When applied to quantitative data, it groups the values into ranges, and the height of each bar shows the frequency of each range.

We can generate a histogram plot of the calories values in the cereal dataframe, which is quantitative. This will enable us to see the various values of calories and how many times they occur.

To make a histogram, we use mark_bar().

In the encode() verb, we specify the x-axis as calories and use the argument bin=True. We assign the y-axis as count():Q to get the number of cereals that have values within each of the ranges.

This is the same count() argument we use in Module 1 when we made bar charts.

Bins

chart6 = alt.Chart(cereal, width=500, height=300).mark_bar().encode(
                   x=alt.X('calories:Q', bin=alt.Bin(maxbins=20)), # set max number of bins to 20
                   y=alt.Y('count():Q')
         ).properties(title="Histogram of cereal calorie content with bins = 20")
chart6

We have the ability to change the number of bars (bins) in our histogram by using the bin argument and the alt.Bin() verb.

Within alt.Bin(), we can specify maxbins which is the maximum allowed number of bins in our plot.

This may be useful when viewing a dataset with lots of different values.

Having control over the number of bins in a histogram can help to make visualization easier to extract insights from.

Here, we set the number of max bins in the plot to 20 by setting bin=alt.Bin(maxbins=20) inside alt.X().

When plotting with Altair, the x and y axis are labelled with the default column names.

This may not always be ideal since column names may not always be informative. In this plot, the x axis label calories (binned) is a little messy.

Luckily Altair allows us to customize our axis labels.

chart7 = alt.Chart(cereal, width=500, height=300).mark_bar().encode(
                   x=alt.X('calories:Q', bin=alt.Bin(maxbins=20), title="Calorie Content"), # use alt.X() to label the x-axis
                   y=alt.Y('count():Q', title="Number of Cereals")                          # use alt.Y() to label the y-axis
        ).properties(title="Histogram plot of cereal calorie content")
chart7

We can change these axis labels using the title="" argument within the respective alt.X() and alt.Y() verbs that we talked about earlier.

This is a big help for the clarity of our analysis.

mfr_mean = cereal.groupby(by='mfr').mean(numeric_only=True)
mfr_mean

	protein	fat	sodium	fiber	...	shelf	weight	cups	rating
mfr
A	4.000000	1.000000	0.000000	0.000000	...	2.000000	1.000000	1.000000	54.850917
G	2.318182	1.363636	200.454545	1.272727	...	2.136364	1.049091	0.875000	34.485852
K	2.652174	0.608696	174.782609	2.739130	...	2.347826	1.077826	0.796087	44.038462
...	...	...	...	...	...	...	...	...	...
P	2.444444	0.888889	146.111111	2.777778	...	2.444444	1.064444	0.714444	41.705744
Q	2.625000	1.750000	92.500000	1.337500	...	2.375000	0.875000	0.823750	42.915990
R	2.500000	1.250000	198.125000	1.875000	...	2.000000	1.000000	0.871250	41.542997

7 rows × 12 columns

In the previous slide deck, we asked the following question regarding our cereal data:

Which manufacturer has the highest mean sugar content?

A nice way of answering this would be to plot the results using a bar chart!

Before doing this, we need a few more tricks.

We can start using the mean statistics we calculated from the groupby(by='mfr') object from the last section.

Here, we seem to have lost our index column of numbers that we usually have. It also appears that mfr has now moved to the left of the dataframe with its label mfr lower than the other column labels.

This is because when you apply groupby() to a column, the grouping column becomes the new dataframe index.

Although this is a useful feature in many cases, Altair cannot access the index column.

mfr_mean

	protein	fat	sodium	fiber	...	shelf	weight	cups	rating
mfr
A	4.000000	1.000000	0.000000	0.000000	...	2.000000	1.000000	1.000000	54.850917
G	2.318182	1.363636	200.454545	1.272727	...	2.136364	1.049091	0.875000	34.485852
K	2.652174	0.608696	174.782609	2.739130	...	2.347826	1.077826	0.796087	44.038462
...	...	...	...	...	...	...	...	...	...
P	2.444444	0.888889	146.111111	2.777778	...	2.444444	1.064444	0.714444	41.705744
Q	2.625000	1.750000	92.500000	1.337500	...	2.375000	0.875000	0.823750	42.915990
R	2.500000	1.250000	198.125000	1.875000	...	2.000000	1.000000	0.871250	41.542997

7 rows × 12 columns

mfr_mean = mfr_mean.reset_index()
mfr_mean

	mfr	protein	fat	sodium	...	shelf	weight	cups	rating
0	A	4.000000	1.000000	0.000000	...	2.000000	1.000000	1.000000	54.850917
1	G	2.318182	1.363636	200.454545	...	2.136364	1.049091	0.875000	34.485852
2	K	2.652174	0.608696	174.782609	...	2.347826	1.077826	0.796087	44.038462
...	...	...	...	...	...	...	...	...	...
4	P	2.444444	0.888889	146.111111	...	2.444444	1.064444	0.714444	41.705744
5	Q	2.625000	1.750000	92.500000	...	2.375000	0.875000	0.823750	42.915990
6	R	2.500000	1.250000	198.125000	...	2.000000	1.000000	0.871250	41.542997

7 rows × 13 columns

To deal with this, we use reset_index() which will convert mfr to a regular column again.

We can see that mfr column has now moved right and our index column of integers has returned on the left!

chart8 = alt.Chart(mfr_mean, width=500, height=300).mark_bar().encode(
                   x=alt.X('mfr:N', title="Manufacturer"),
                   y=alt.Y('sugars:Q', title="Mean sugar content")
         ).properties(title="Bar plot of manufacturers mean sugar content")
chart8

Now that we have our mfr_mean in the correct format, we can proceed.

Using Altair we can plot the mfr column on the x axis which we’ve identified to contain nominal values and sugars which we agreed was a quantitative value on the y axis. (Also, let’s not forget our title!)

1. Groupby object and calculated the mean
2. Reset index
3. Plot using Altair

Let’s go through the steps that were needed to make the plot in the previous slide.

First, we created a groupby object and calculated the mean for each column in the resulting dataframe.
Second, since .groupby() made mfr the new index, we had to use reset_index() to make mfr a regular column again. And finally, we generated a bar plot using Altair.

Sorting

chart9 = alt.Chart(mfr_mean, width=500, height=300).mark_bar().encode(
                   x=alt.X('mfr:N', sort="y", title="Manufacturer"),  # use sort="y" to sort in ascending order
                   y=alt.Y('sugars:Q', title="Mean sugar content")
        ).properties(title="Bar plot of manufacturers mean sugar content in ascending order")
chart9

Sometimes sorting a dataframe by quantity helps us obtain insights more easily.

For example, if we sorted the mean sugar content for the manufacturers before generating the previous plot, it would be easier to identify which manufacturer produces cereals with the highest mean sugar content.

Altair allows us to sort a column while plotting.

Sorting can be done on either the x or y axis using the sort= in the alt.X or alt.Y verb.

The sort argument takes in either x or y to specify which axis to sort by.

Here we are sorting in ascending order of which manufacturers have the largest mean sugar content.

This plot shows us immediately that manufacturer P has the highest mean cereal sugar content.

chart10 = alt.Chart(mfr_mean, width=500, height=300).mark_bar().encode(
    x=alt.X('mfr:N', sort="-y", title="Manufacturer"),  # use sort="-y" to sort in descending order
    y=alt.Y('sugars:Q', title="Mean sugar content")
).properties(title="Bar plot of manufacturers mean sugar content sorted in descending order")
chart10

To generate a bar plot of mean calorie content sorted in descending order, we recycle the code from the previous slide.

This time, we add -y in the sort argument to specify that we would like to sort the y variable in descending order.

If you enjoyed this part of the module and you wish to learn more advanced visualizations using Altair, take a look at our
Data Visualization course

Let’s apply what we learned!