Setting category variables

Working with Categorical Data in Python

Kasey Jones

Research Data Scientist

New dataset: adoptable dogs

dogs.info()

RangeIndex: 2937 entries, 0 to 2936, Data columns (total 19 columns):
 #   Column             Non-Null Count  Dtype  
 --  ------             --------------  -----  
 0   ID                 2937 non-null   int64  
 ...
 8   color              2937 non-null   object 
 9   coat               2937 non-null   object 
 ...
 17  get_along_cats     431 non-null    object 
 18  keep_in            1916 non-null   object 
dtypes: float64(1), int64(1), object(17)
memory usage: 436.1+ KB
...

¹ https://www.kaggle.com/jmolitoris/adoptable-dogs

A dog's coat

dogs["coat"] = dogs["coat"].astype("category")
dogs["coat"].value_counts(dropna=False)

short         1972
medium         565
wirehaired     220
long           180
Name: coat, dtype: int64

The .cat accessor object

Series.cat.method_name

Common parameters:

new_categories: a list of categories
inplace: Boolean - whether or not the update should overwrite the Series
ordered: Boolean - whether or not the categorical is treated as an ordered categorical

Setting Series categories

Set categories:

dogs["coat"] = dogs["coat"].cat.set_categories(
  new_categories=["short", "medium", "long"]
)

Check value counts:

dogs["coat"].value_counts(dropna=False)

short     1972
medium     565
NaN        220
long       180

Setting order

dogs["coat"] = dogs["coat"].cat.set_categories(
  new_categories=["short", "medium", "long"],
  ordered=True
)
dogs["coat"].head(3)

0     short
1     short
2     short
Name: coat, dtype: category
Categories (3, object): ['short' < 'medium' < 'long']

Missing categories

dogs["likes_people"].value_counts(dropna=False)

yes    1991
NaN     938
no        8

A NaN could mean:

Truly unknown (we didn't check)
Not sure (dog likes "some" people)

Adding categories

Add categories

dogs["likes_people"] = dogs["likes_people"].astype("category")
dogs["likes_people"] = dogs["likes_people"].cat.add_categories(
  new_categories=["did not check", "could not tell"]
)

Check categories:

dogs["likes_people"].cat.categories

Index(['no', 'yes', 'did not check', 'could not tell'], dtype='object')

New categories

dogs["likes_people"].value_counts(dropna=False)

yes               1991
NaN                938
no                   8
could not tell       0
did not check        0

Removing categories

dogs["coat"] = dogs["coat"].astype("category")
dogs["coat"] = dogs["coat"].cat.remove_categories(removals=["wirehaired"])

Check the categories:

dogs["coat"].cat.categories

Index(['long', 'medium', 'short'], dtype='object')

Methods recap

Setting: cat.set_categories()
- Can be used to set the order of categories
- All values not specified in this method are dropped
Adding: cat.add_categories()
- Does not change the value of any data in the DataFrame
- Categories not listed in this method are left alone
Removing: cat.remove_categories()
- Values matching categories listed are set to NaN

Practice updating categories

Working with Categorical Data in Python