Categorical variables are non-numeric and cannot be directly used in linear regression. They need to be transformed into a format suitable for numerical analysis. Try convert them into numerical values using one hot encoding using sklearn. Identify which features in your dataset are categorical. These could be variables like "Gender," "City," or "Product Type." One-hot encoding is a method to represent categorical variables as binary vectors (0s and 1s). For each category in a variable, a new binary column is created. This is how you can do it using sklearn. 

import pandas as pd
from sklearn.preprocessing import OneHotEncoder

# Create a sample DataFrame
data = {'City': ['New York', 'San Francisco', 'Chicago']}
df = pd.DataFrame(data)

# Apply One-Hot Encoding
encoder = OneHotEncoder()
encoded_data = encoder.fit_transform(df[['City']])

After one-hot encoding, you'll get a sparse matrix with binary values representing the presence of each category. In this example, 'New York,' 'San Francisco,' and 'Chicago' would each have their binary column. After one-hot encoding, you can concatenate the new binary columns with the original dataset. 

# Concatenate with original DataFrame
df_encoded = pd.concat([df, pd.DataFrame(encoded_data.toarray(), columns=encoder.get_feature_names_out(['City']))], axis=1)

You can drop the original categorical column as it is no longer needed.

When using one-hot encoding, be mindful of the dummy variable trap. This occurs when the values of one variable can be predicted from the values of the others. To avoid this, drop one of the binary columns for each categorical variable. When working with more complex workflows, consider using scikit-learn pipelines, which streamline the process of data preprocessing and modeling.