• API Reference - StringMismatchComparison

String Mismatch Comparison

This page provides an overview for using and understanding the “String Mismatch Comparison” check.

Structure:

• What is the purpose of the check?

• Run check

• Define a condition

What is the purpose of the check?

The check compares the same categorical column within train and test and checks whether there are variants of similar strings that exists only in test and not in train. Finding those mismatches is helpful to prevent errors when inferring on the test data. For example, in train data we have category ‘New York’, and in our test data we have ‘new york’. We would like to be acknowledged that the test data contain a new variant of the train data, so we can address the problem.

How String Mismatch Defined?

To recognize string mismatch, we transform each string to it’s base form. The base form is the string with only its alphanumeric characters in lowercase. (For example “Cat-9?!” base form is “cat9”). If two strings have the same base form, they are considered to be the same.

import pandas as pd

Run the Check

from deepchecks.tabular.checks import StringMismatchComparison

data = {'col1': ['Deep', 'deep', 'deep!!!', 'earth', 'foo', 'bar', 'foo?']}
compared_data = {'col1': ['Deep', 'deep', '$deeP$', 'earth', 'foo', 'bar', 'foo?', '?deep']}

check = StringMismatchComparison()
result = check.run(pd.DataFrame(data=data), pd.DataFrame(data=compared_data))
result

Define a Condition

check = StringMismatchComparison().add_condition_no_new_variants()
result = check.run(pd.DataFrame(data=data), pd.DataFrame(data=compared_data))
result.show(show_additional_outputs=False)

previous

New Label

next

Train Test Feature Drift