Valuable Data Analysis with Pandas Value Counts

The data

In the examples shown in this article, I will be using a data set taken from the Kaggle website. It is designed for a machine learning classification task and contains information about medical appointments and a target variable which denotes whether or not the patient showed up to their appointment.

It can be downloaded here .

In the code below I have imported the data and the libraries that I will be using throughout the article.

import pandas as pdimport matplotlib.pyplot as plt
%matplotlib inlinedata = pd.read_csv('KaggleV2-May-2016.csv')
data.head()

Basic counts

The value_counts() function can be used in the following way to get a count of unique values for one column in the data set. The code below gives a count of each value in the Gender column.

data['Gender'].value_counts()

To sort values in ascending or descending order we can use the sort argument. In the code below I have added sort=True to display the counts in the Age column in descending order.

data['Age'].value_counts(sort=True)

Combine with groupby()

The value_counts function can be combined with other Panadas functions for richer analysis techniques. One example is to combine with the groupby() function. In the below example I am counting values in the Gender column and applying groupby() to further understand the number of no-shows in each group.

data['No-show'].groupby(data['Gender']).value_counts(sort=True)

Normalize

In the above example displaying the absolute values does not easily enable us to understand the differences between the two groups. A better solution would be to show the relative frequencies of the unique values in each group.

We can add the normalize argument to value_counts() to display the values in this way.

data['No-show'].groupby(data['Gender']).value_counts(normalize=True)

Binning

For columns where there are a large number of unique values the output of the value_counts() function is not always particularly useful. A good example of this would be the Age column which we displayed value counts for earlier in this post.

Fortunately value_counts() has a bins argument. This parameter allows us to specificy the number of bins (or groups we want to split the data into) as an integer. In the example below I have added bins=5 to split the Age counts into 5 groups. We now have a count of values in each of these bins.

data['Age'].value_counts(bins=5)

Once again showing absolute numbers is not particularly useful so let’s add the normalize=True argument as well. Now we have a useful piece of analysis.

data['Age'].value_counts(bins=5, normalize=True)

Combine with nlargest()

There are other columns in our data set which have a large number of unique values where binning is still not going to provide us with a useful piece of analysis. A good example of this would be the Neighbourhood column.

If we simply run value_counts() against this we get an output that is not particularly insightful.

data['Neighbourhood'].value_counts(sort=True)

A better way to display this might be to view the top 10 neighbourhoods. We can do this by combining with another Pandas function called nlargest() as shown below.

data['Neighbourhood'].value_counts(sort=True).nlargest(10)

We can also use nsmallest() to display the bottom 10 neighbourhoods which might also prove useful.

data['Neighbourhood'].value_counts(sort=True).nsmallest(10)

Plotting

Another handy combination is the Pandas plotting functionality together with value_counts(). Having the ability to display the analyses we get from value_counts() as visualisations can make it far easier to view trends and patterns.

We can display all of the above examples and more with most plot types available in the Pandas library. A full list of available options can be found here .

Let’s look a few examples.

We can use a bar plot to view the top 10 neighbourhoods.

data['Neighbourhood'].value_counts(sort=True).nlargest(10).plot.bar()

We can make a pie chart to better visualise the Gender column.

data['Gender'].value_counts().plot.pie()