How to Create Columns in Pandas

Creating new columns is one of the most important skills in data analysis. Whether you're computing a derived metric, binning continuous data into categories, or preparing features for a machine learning model — you'll be adding columns to a DataFrame constantly.

In this article you'll learn three things:

1. How to create and assign values to a column
2. How to create a categorical column with pd.cut()
3. How to create one-hot encodings with pd.get_dummies()

1. The Dataset

We'll use a petrol station dataset with fuel transactions across Australia. Each row has a station, fuel type, litres sold, price per litre, and the state. Some values are intentionally missing.

2. Create a Column with Bracket Assignment

The simplest way to create a new column is bracket assignment. Let's add a revenue column initialised to zero:

You can also use loc to achieve the same thing. This is useful when working with filtered DataFrames to avoid the SettingWithCopyWarning:

3. Populate with Random Values

Zeros aren't very interesting. Let's populate the discount column with random values between 0 and 15 (cents off per litre) using np.random.randint():

4. Compute a Derived Column

Now let's compute the actual revenue column. Revenue = litres x price per litre. But there's a catch — some rows have NaN in litres or price_per_litre. If you multiply a number by NaN, the result is NaN. Use fillna(0) first:

Let's also compute the discounted price by subtracting the discount (in cents) from the price per litre:

Let's clean up the extra columns and keep the ones we need going forward:

5. Create Categorical Columns with pd.cut()

You have a continuous revenue column. What if the business wants to categorise transactions into groups — say Low, Medium, and High revenue? This is where pd.cut() comes in. It takes a continuous column and bins it into discrete categories.

First, let's calculate the bins dynamically from the data:

Now create labels for each bin range and apply pd.cut():

Now you can group by the category to see how many transactions fall into each bucket:

6. One-Hot Encoding with pd.get_dummies()

Categorical columns are great for analysis, but machine learning models typically need numerical inputs. One-hot encoding converts each category into its own binary column (1 or 0). Use pd.get_dummies() to do this in one line:

The dummy_na=True parameter adds a column for NaN values — useful if missing data is meaningful. Now merge the dummies back into the original DataFrame:

The DataFrame now has binary indicator columns ready for any machine learning pipeline or statistical model.

Summary

You've learned three essential column creation techniques:

1. Direct assignment — df['col'] = value or df.loc[:, 'col'] = value to create new columns with constants, random values, or computed expressions
2. Categorical binning — pd.cut() to bin continuous values into labelled categories with dynamic or manual bins
3. One-hot encoding — pd.get_dummies() to convert categorical columns into binary indicators for ML models

Try editing the code blocks above — change the number of bins, use fuel_type instead of revenue_category for one-hot encoding, or compute new derived columns like revenue per litre.