Feature encoding
Prepare categorical features for a machine-learning model.
Pandas · Function
pd.get_dummies
Converts categorical values into 0/1 indicator columns. DataFrame inputs can selectively encode columns and retain other columns.
How it works
pd.get_dummies converts categorical values into indicator columns containing 0 or 1. For a DataFrame, selected columns—or, when columns is omitted, columns with object, string, or category dtype—are expanded and named using the source column and category value.
Syntax
pandas.get_dummies(data, prefix=None, prefix_sep='_', dummy_na=False, columns=None, sparse=False, drop_first=False, dtype=None)Parameters
| Name | Type | Description |
|---|---|---|
data | array-like, Series, or DataFrame | Array-like, Series, or DataFrame containing the data to encode. |
prefix | str, list of str, or dict of str, default None | String, list of strings, or mapping used to prefix DataFrame column names; default None. |
prefix_sep | str, default '_' | Separator used between a prefix and category value; default '_'. |
dummy_na | bool, default False | Whether to add an indicator for missing values; default False. |
columns | list-like, default None | DataFrame column names to encode; default None. |
sparse | bool, default False | Whether generated columns use SparseArray backing; default False. |
drop_first | bool, default False | Whether to omit the first category level; default False. |
dtype | dtype, default None | Data type for the generated indicators; default None. |
Returns
DataFrame of dummy-coded data.
Key points
- The function transforms categorical data into numeric indicator variables.
- A categorical variable receives one binary column per distinct value unless a level is dropped.
- Generated columns use category values in their names; DataFrame inputs also include the original column name.
- prefix customizes the text prepended to encoded DataFrame column names.
- Use columns to select which DataFrame fields should be converted.
- sparse=True stores the encoded columns with sparse-array backing rather than regular dense storage.
Common mistakes
- With dummy_na=False, missing values do not receive a separate indicator and are represented by zeros.
- Pass one dtype for all generated indicator columns; different dtypes per output column are not supported.
Example: Encode a DataFrame column
import pandas as pd
df = pd.DataFrame({"color": ["red", "blue", "red"], "value": [1, 2, 3]})
encoded = pd.get_dummies(df, columns=["color"], dtype=int)
print(encoded)Encodes the selected categorical column as integer indicator columns while retaining the numeric column.
When to use it
Data transformation
Turn labeled categories into analysis-ready indicator columns.
Missing-value indicators
Represent missingness explicitly as a model feature.
Interview questions
QWhat does pd.get_dummies do?
It creates one indicator column for each distinct category in the input. For DataFrame input, generated names combine the original column name with the category value.
Follow-up: How would you avoid encoding every eligible column? Pass an explicit list of column names through columns.
QWhich DataFrame columns are encoded?
By default, columns is None, so object, string, and category columns are converted. Supplying columns restricts encoding to the named DataFrame columns.
Follow-up: What happens to columns not listed in columns? They remain in the returned DataFrame unchanged.
QWhat is the effect of drop_first?
drop_first=True removes the first category level, producing k−1 indicator columns for k levels. This can remove a redundant reference category in models.
Follow-up: How are missing values handled when dummy_na=True? A separate indicator column is added for missing values.
Related
pandas.from_dummiespandas.Categoricalpandas.Series.str.get_dummies