Label encoding

Làm việc với dữ liệu phân loại trong Python

Kasey Jones

Research Data Scientist

What is label encoding?

The basics:

Codes each category as an integer from 0 through n - 1, where n is the number of categories
A -1 code is reserved for any missing values
Can save on memory
Often used in surveys

The drawback:

Is not the best encoding method for machine learning (see next lesson)

Creating codes

Convert to categorical and sort by manufacturer name

used_cars['manufacturer_name'] = used_cars['manufacturer_name'].astype("category")

Use .cat.codes

used_cars['manufacturer_code'] = used_cars['manufacturer_name'].cat.codes

Check output

print(used_cars[['manufacturer_name', 'manufacturer_code']])

      manufacturer_name  manufacturer_code
0                Subaru                 45
1                Subaru                 45
2                Subaru                 45
...                 ...                ...
38526          Chrysler                  8
38527          Chrysler                  8

Code books / data dictionaries

An example code book from the American Housing Survey.

¹ https://www.census.gov/data-tools/demo/codebook/ahs/ahsdict.html

Creating a code book

codes = used_cars['manufacturer_name'].cat.codes
categories = used_cars['manufacturer_name']

name_map = dict(zip(codes, categories))

print(name_map)

{45: 'Subaru',
 24: 'LADA',
 12: 'Dodge',
 ...
}

Using a code book

Creating the codes:

used_cars['manufacturer_code'] = used_cars['manufacturer_name'].cat.codes

Reverting to previous values:

used_cars['manufacturer_code'].map(name_map)

0        Acura
1        Acura
2        Acura
...

¹ https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.Series.map.html

Boolean coding

Find all body types that have "van" in them:

# Code from previous lesson:
used_cars["body_type"].str.contains("van", regex=False)

Create a boolean coding:

used_cars["van_code"] = np.where(
  used_cars["body_type"].str.contains("van", regex=False), 1, 0)

used_cars["van_code"].value_counts()

0    34115
1     4416
Name: van_code, dtype: int64

Encoding practice

Làm việc với dữ liệu phân loại trong Python