Introduction to Grouping

Concept

The ability to group or segment DataFrames by one or more columns is one of the key features of any data analysis application. Therefore it would most likely show up in a data analysis interview or task.

The idea is to divide a DataFrame into multiple groups to analyze each group separately.

Syntax

As a reminder, the syntax is as simple as df.groupby(<col_name>) or, in the case of grouping by multiple columns, df.groupby([<col1>, <col2>, ..]).

Operations such as aggregations and apply functions can be applied on DataFrameGroupBy objects which can be reset to a normal DataFrame using reset_index()

Travel dataset

Idea

In your travel dataset, it makes sense that not all travelers exhibit the same behavior. Certain nationalities might like to travel to specific destinations, while other nationalities might prefer other destinations.

In this case, your grouping column will be nationality, and you will be analyzing the destination distribution per nationality. You might want the most common destinations or the count of unique destinations. This can help in developing marketing strategies for nation-destination combinations.

Another example is that certain types of tour packages might be more interesting in specific weather conditions; for example, cruises in the summer.

In this case, your grouping column will be tour_type, and you will be analyzing the season distribution.

In interviews, it is important to understand the business domain and use-case before answering a question. So, it’s likely that you’ll be using either grouping or filtering on the most important attributes representing the business core models.

In the example here, the core models are the traveler and the trip. It’s important to pause here and reflect on what attributes would likely affect the behaviors of travelers. Here, you discussed nationality, destination, tour_type, and season. Can you come up with other core attributes?

Interviews

You can be asked open-ended questions interviews, such as the following:

1. How would you create a semi-personalized newsletter to send to previous travelers based on their demographics, characteristics, or previous trips?

One solution here would be to segment or group the previous travelers by their most distinguishing characteristics. Then you can send them suggestions for other trips they are likely to be interested in.

Medical dataset

Idea

In your medical dataset, patients have different health conditions, and receive different treatments. Patients with heart disease might show different symptoms, take specific medications, and mostly belong to a specific demographic.

You might want to see the correlation of diabetes condition with other demographic information and the most commonly received treatments. This could help in awareness campaigns for at-risk patient categories.

In this case, one obvious grouping column is the health_condition, and based on it you could analyze the age, gender, weight, treatments, and so on.

Interviews

You can be asked open-ended questions, such as:

1. What is the top health condition affecting every age range (children, teens, young adults, adults, and seniors)?

Here, one solution would be to segment or group the patients by age range, and show the top 2–3 health conditions each group is suffering from.

Before you move to the challenges

1. Make sure you know how to groupby using single/multiple columns.
2. Make sure you know how to apply the agg function on DataFrameGroupBy objects. This should include applying on different columns, and the different possible functions to apply.
3. Make sure you know how to apply reset_index to return a DataFrame.