Pandas: fillna only numeric (int or float) columns

You can select numeric columns and then fillna E.g:

import pandas as pd

df = pd.DataFrame({'a': [1, None] * 3,
                    'b': [True, None] * 3,
                  'c': [1.0, None] * 3})

# select numeric columns
numeric_columns = df.select_dtypes(include=['number']).columns

# fill -1 to all NaN 
df[numeric_columns] = df[numeric_columns].fillna(-1)

# print
print(df)

It's an old question, however, I discovered that individually filling the columns is faster than the currently chosen answer:

def func(df, value):
    df = df.copy()
    for col in df:
        # select only integer or float dtypes
        if df[col].dtype in ("int", "float"):
            df[col] = df[col].fillna(value)
    return df

 func(df, value=-1) # or df.pipe(func, value=-1)

      a      b        c
0    1.0    True     1.0
1   -1.0    None    -1.0
2    1.0    True     1.0
3   -1.0    None    -1.0
4    1.0    True     1.0
5   -1.0    None    -1.0

comparing speeds the loop returns 470 µs ± 12.1 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each), while the accepted answer returns 1.57 ms ± 26.3 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each).

If the dataframe size is increased to 60,000 rows: pd.concat([df]*10_000, ignore_index=True), the loop returns 1.48 ms ± 79.2 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each) while the chosen answer returns 2.47 ms ± 140 µs per loop (mean ± std. dev. of 7 runs, 100 loops each).

For both instances, the loop is significantly faster than the chosen answer. Also, your mileage may vary. Just some food for thought, especially when trying to wring out more performance.