Python Pandas - Changing some column types to categories

Sometimes, you just have to use a for-loop:

for col in ['parks', 'playgrounds', 'sports', 'roading']:
    public[col] = public[col].astype('category')

You can use the pandas.DataFrame.apply method along with a lambda expression to solve this. In your example you could use

df[['parks', 'playgrounds', 'sports']].apply(lambda x: x.astype('category'))

I don't know of a way to execute this inplace, so typically I'll end up with something like this:

df[df.select_dtypes(['object']).columns] = df.select_dtypes(['object']).apply(lambda x: x.astype('category'))

Obviously you can replace .select_dtypes with explicit column names if you don't want to select all of a certain datatype (although in your example it seems like you wanted all object types).

No need for loops, Pandas can do it directly now, just pass a list of columns you want to convert and Pandas will convert them all.

cols = ['parks', 'playgrounds', 'sports', 'roading']
public[cols] = public[cols].astype('category')

df = pd.DataFrame({'a': ['a', 'b', 'c'], 'b': ['c', 'd', 'e']})

>>     a  b
>>  0  a  c
>>  1  b  d
>>  2  c  e

df.dtypes
>> a    object
>> b    object
>> dtype: object

df[df.columns] = df[df.columns].astype('category')
df.dtypes
>> a    category
>> b    category
>> dtype: object

As of pandas 0.19.0, What's New describes that read_csv supports parsing Categorical columns directly. This answer applies only if you're starting from read_csv otherwise, I think unutbu's answer is still best. Example on 10,000 records:

import pandas as pd
import numpy as np

# Generate random data, four category-like columns, two int columns
N=10000
categories = pd.DataFrame({
            'parks' : np.random.choice(['strongly agree','agree', 'disagree'], size=N),
            'playgrounds' : np.random.choice(['strongly agree','agree', 'disagree'], size=N),
            'sports' : np.random.choice(['important', 'very important', 'not important'], size=N),
            'roading' : np.random.choice(['important', 'very important', 'not important'], size=N),
            'resident' : np.random.choice([1, 2, 3], size=N),
            'children' : np.random.choice([0, 1, 2, 3], size=N)
                       })
categories.to_csv('categories_large.csv', index=False)

<0.19.0 (or >=19.0 without specifying dtype)

pd.read_csv('categories_large.csv').dtypes # inspect default dtypes

children        int64
parks          object
playgrounds    object
resident        int64
roading        object
sports         object
dtype: object

>=0.19.0

For mixed dtypes parsing as Categorical can be implemented by passing a dictionary dtype={'colname' : 'category', ...} in read_csv.

pd.read_csv('categories_large.csv', dtype={'parks': 'category',
                                           'playgrounds': 'category',
                                           'sports': 'category',
                                           'roading': 'category'}).dtypes
children          int64
parks          category
playgrounds    category
resident          int64
roading        category
sports         category
dtype: object

Performance

A slight speed-up (local jupyter notebook), as mentioned in the release notes.

# unutbu's answer
%%timeit
public = pd.read_csv('categories_large.csv')
for col in ['parks', 'playgrounds', 'sports', 'roading']:
    public[col] = public[col].astype('category')
10 loops, best of 3: 20.1 ms per loop

# parsed during read_csv
%%timeit
category_cols = {item: 'category' for item in ['parks', 'playgrounds', 'sports', 'roading']}
public = pd.read_csv('categories_large.csv', dtype=category_cols)
100 loops, best of 3: 14.3 ms per loop