Altcademy - a Forbes magazine logo Best Coding Bootcamp 2023

How to one hot encode a column in Pandas

Understanding One Hot Encoding

Before diving into the technical aspects of one hot encoding in Pandas, let's grasp the concept with a simple analogy. Imagine you have a collection of colored balls: red, green, and blue. If you wanted to organize them in a way that a computer could easily understand which color is present, you could create a separate box for each color. Now, if you place a ball into the corresponding box, you can represent the presence of a color with a simple 'yes' or 'no' (or in computer terms, '1' or '0') for each box. This is, in essence, what one hot encoding does with categorical data.

Categorical data refers to variables that contain label values rather than numeric values. The number of possible values is often limited to a fixed set, like our example with colored balls. One hot encoding is a process by which categorical variables are converted into a form that could be provided to machine learning algorithms to do a better job in predictions.

Why One Hot Encode?

In many machine learning scenarios, you'll deal with data that is categorical. If you feed this data directly into a model, it may misinterpret the categorical data as some sort of rank or order (which blue is greater than green?), which isn't usually the case. One hot encoding transforms the categorical data into a format that prevents this issue by creating a binary column for each category.

Getting Started with Pandas for One Hot Encoding

Pandas is a powerful Python library for data manipulation and analysis. It provides numerous functions to deal with different data types, including tools for one hot encoding. To get started with one hot encoding in Pandas, you first need a dataset that contains categorical data.

Imagine we have a dataset of pets with a column for species which includes categories like 'Dog', 'Cat', and 'Bird'. We'll use this as our example dataset to demonstrate one hot encoding.

import pandas as pd

# Sample dataset of pets
data = {'Pet': ['Dog', 'Cat', 'Bird', 'Dog', 'Cat']}
df = pd.DataFrame(data)

The get_dummies Function in Pandas

Pandas simplifies the one hot encoding process with a function called get_dummies. This function automatically converts all categorical variables in a DataFrame to one hot encoded vectors.

Here's how you can use get_dummies to one hot encode the 'Pet' column:

# One hot encoding the 'Pet' column
encoded_df = pd.get_dummies(df, columns=['Pet'])

The resulting DataFrame encoded_df will have additional columns for each unique value in the 'Pet' column, with binary indicators showing the presence of each category.

Understanding the Output

The output DataFrame from the get_dummies function will look something like this:

   Pet_Cat  Pet_Dog  Pet_Bird
0        0        1         0
1        1        0         0
2        0        0         1
3        0        1         0
4        1        0         0

Each row now has a set of columns corresponding to the possible categories. A '1' indicates the presence of that category and a '0' indicates its absence.

Dealing with Unseen Categories

What happens if new data comes in with categories that weren't present in the original dataset? This is an important consideration, as machine learning models trained on the original encoded dataset may not know how to handle these new categories.

To address this, you can create a function that aligns the new data with the original dataset's columns:

def encode_and_align(df, new_data):
    # One hot encode the new data
    new_encoded = pd.get_dummies(new_data)

    # Align the new data with the original DataFrame's columns
    final_encoded = new_encoded.reindex(columns=df.columns, fill_value=0)

    return final_encoded

# Sample new data with unseen category 'Fish'
new_data = pd.DataFrame({'Pet': ['Fish']})
new_encoded_data = encode_and_align(encoded_df, new_data)

This way, even if the new data contains categories like 'Fish' that weren't in the original dataset, it can be properly processed without causing errors in the model.

Preserving Column Order

When you one hot encode data, the order of the new columns is typically alphabetical. However, you may want to preserve the order of categories as they appeared in the original dataset. To do this, you can specify the columns parameter in the get_dummies function:

# One hot encoding with column order preserved
encoded_df_ordered = pd.get_dummies(df['Pet'], prefix='Pet', columns=df['Pet'].unique())

Handling Missing Values

Sometimes, your categorical data might have missing values. Pandas get_dummies function automatically handles missing values by not creating a '1' in any of the one hot encoded columns for that row.

However, you might want to explicitly mark missing values. One way to do this is to first fill missing values with a placeholder category, and then apply one hot encoding:

df['Pet'].fillna('Unknown', inplace=True)
encoded_df_with_missing = pd.get_dummies(df, columns=['Pet'])

When Not to Use One Hot Encoding

One hot encoding can increase the dimensionality of your dataset significantly if you have a lot of unique categories. This can lead to a problem known as the "curse of dimensionality," where the feature space becomes so large that the model's performance may actually degrade. In such cases, other encoding techniques like label encoding or feature hashing may be more appropriate.


One hot encoding is a vital preprocessing step in the journey of a machine learning model. It's like giving your model a pair of glasses so it can see the differences between categories clearly and without confusion. With Pandas, the process is straightforward and efficient, allowing you to prepare your data for better predictions.

By understanding and applying one hot encoding, you're taking a significant step towards making your data more comprehensible and actionable for machine learning algorithms. Remember, the key is to ensure your data accurately represents the real-world scenarios your model will face, without introducing unnecessary complexity. Keep practicing with different datasets, and soon, one hot encoding will be as intuitive as sorting those colored balls into their respective boxes.