Signature
DataFrame.fillna(value=None, *, method=None, axis=None, inplace=False)What it does
fillna returns a copy with missing values replaced. Prefer fitting the replacement value on training data, especially when the value is a mean or median, so the test set does not influence preprocessing.
Parameters
valuescalar, dict, Series, or DataFrame- Replacement value or per-column mapping.
methodstr, optionaloptional- Fill forward or backward instead of using a value.
inplacebool, optionaloptional- Mutate the DataFrame instead of returning a copy.
Returns
DataFrame or None
Example
pythonDataFrame.fillna example
import pandas as pd
import numpy as np
df = pd.DataFrame({"score": [10.0, np.nan, 14.0]})
print(df.fillna({"score": df["score"].mean()}))Output
score 0 10.0 1 12.0 2 14.0
Watch out for
- Computing the fill value on all rows leaks test information into training.
- A single constant can hide why a value was missing; consider a missing indicator.