Skip to main content
Version: 1.2.1

Apply Conditional Expectations to specific rows within a Batch

By default Expectations apply to the entire dataset retrieved in a Batch. However, sometimes an Expectation is not relevant for every row and validating every row could cause false positives or false negatives in the Validation Results.

For example, you may define an Expectation that a column specifying the country of origin of a product should not be null. If that Expectation is only relevant when the product is a foreign import then applying the Expectation to every row in the Batch could result in a large number of false negatives when the country of origin column is null for products produced by local industry.

To solve this issue GX allows you to define Conditional Expectations that only apply to a subset of the data retrieved in a Batch.

Create a Conditional Expectation​

Great Expectations lets you express Conditional Expectations with a row_condition argument that can be passed to all Expectations that evaluate rows within a Dataset. The row_condition argument should be a boolean expression string. The Conditional Expectation will validate rows that result in the row_condition string being True. When the row_condition string evaluates as False, the row in question will not be validated by the Expectation.

Prerequisites​

Procedure​

In this procedure, your Data Context is assumed to be stored in the variable context and your Expectation Suite is assumed to be stored in the variable suite. suite can be a newly created and empty Expectation Suite, or an existing Expectation Suite retrieved from the Data Context.

The data used in the examples for this procedure is passenger data for the Titanic, including what class of ticket the passenger held and whether or not they survived the journey.

  1. Determine the condition_parser for your row_condition.

    The condition_parser defines the syntax of row_condition strings. When implementing conditional Expectations with pandas, this argument must be set to "pandas". When implementing conditional Expectations with Spark or SQLAlchemy, this argument must be set to "great_expectations".

    Conditional Expectations will fail if the Batch they are validating comes from a different type of Data Source than is indicated by the condition_parser argument.

  2. Determine the row_condition expression.

    The row_condition argument should be a boolean expression string which will be evaluated for each row in the Batch the Expectation validates. When the row_condition evaluates as True the row will be included in the Expectation's validations. When the row_condition evaluates as False, the Expectation will be skipped for that row.

    The syntax of the row_condition argument is based on the condition_parser that was previously specified.

    In pandas the row_condition value is passed to pandas.DataFrame.query() before Expectation Validation and the returned rows from the evaluated Batch will be validated by the Conditional Expectation.

  3. Create a Conditional Expectation.

    A Conditional Expectation is created exactly like a regular Expectation, except that the row_condition and condition_parser parameters are provided in addition to the Expectation's other arguments.

    Python
    condition_parser="pandas",
    row_condition='PClass=="1st"',

    Do not use single quotes, newlines, or \n inside the specified row_condition as shown in the following examples:

    Python
    row_condition = "PClass=='1st'"  # Don't do this. Single quotes aren't valid!

    row_condition="""
    PClass=="1st"
    """  # Don't do this.  Newlines and \n aren't valid!

    row_condition = 'PClass=="1st"'  # Do this instead.

    With pandas you can indicate variables from the environment by prefacing them with @. You can also indicate columns with a space in their name by wrapping the name with backticks: `.

    Some examples of valid row_condition values for pandas include:

    Python
    row_condition = '`foo foo`=="bar bar"'  # The value of the column "foo foo" is "bar bar"

    row_condition = 'foo==@bar'  # the value of the foo field is equal to the value of the bar environment variable

    For more information on the syntax accepted by pandas row_condition values see pandas.DataFrame.query.

  4. Optional. Create additional Conditional Expectations.

    Expectations with different conditions are treated as unique even if they are of the same type and apply to the same column within an Expectation Suite. This allows you to create one unconditional Expectation and an arbitrary number of Conditional Expectations (each with a different condition).

    For example, the following code creates a unconditional Expectation that the value of the "Survived" column is either 0 or 1:

    Python
    expectation = gx.expectations.ExpectColumnValuesToBeInSet(
        column="Survived", value_set=[0, 1]
    )

    And this code creates a Conditional version of the same Expectation that specifies the value of the "Survived" column is 1 if the individual was a first class passenger:

    Python
    conditional_expectation = gx.expectations.ExpectColumnValuesToBeInSet(
        column="Survived",
        value_set=[1],
        condition_parser="pandas",
        row_condition='PClass=="1st"',
    )

Data Docs and Conditional Expectations​

Conditional Expectations are displayed differently from standard Expectations in the Data Docs. Each Conditional Expectation is qualified with if 'row_condition_string', then values must be... as shown in the following image:

Image

If 'row_condition_string' is a complex expression, it is split into several components to improve readability.

Scope and limitations​

While conditions can be attached to most Expectations, the following Expectations cannot be conditioned and do not take the row_condition argument:

  • expect_column_to_exist
  • expect_table_columns_to_match_ordered_list
  • expect_table_columns_to_match_set
  • expect_table_column_count_to_be_between
  • expect_table_column_count_to_equal