Skip to main content
Version: 0.14.13

Standard arguments for Expectations

All Expectations return a JSON-serializable dictionary when evaluated, and share four standard (optional) arguments:

  • result_format: Controls what information is returned from the evaluation of the Expectation.
  • include_config: If true, then the Expectation Suite itself is returned as part of the result object.
  • catch_exceptions: If true, execution will not fail if the Expectation encounters an error. Instead, it will return success = False and provide an informative error message.
  • meta: Allows user-supplied meta-data to be stored with an Expectation.

All ColumnMapExpectations also have the following argument:

  • mostly: A special argument that allows for fuzzy validation based on some percentage (available for all column_map_expectations)

result_format

See Result format for more information.

include_config

All Expectations accept a boolean include_config parameter. If true, then the Expectation Suite itself is returned as part of the result object

expect_column_values_to_be_in_set(
"my_var",
['B', 'C', 'D', 'F', 'G', 'H'],
result_format="COMPLETE",
include_config=True,
)
# This returns:
{
'exception_index_list': [0, 10, 11, 12, 13, 14],
'exception_list': ['A', 'E', 'E', 'E', 'E', 'E'],
'expectation_type': 'expect_column_values_to_be_in_set',
'expectation_kwargs': {
'column': 'my_var',
'result_format': 'COMPLETE',
'value_set': ['B', 'C', 'D', 'F', 'G', 'H']
},
'success': False
}

catch_exceptions

All Expectations accept a boolean catch_exceptions parameter. If this parameter is set to True, then Great Expectations will intercept any exceptions so that execution will not fail if the Expectation encounters an error. Instead, if Great Excpectations catches an exception while evaluating an Expectation, the Expectation result will ( in BASIC and SUMMARY modes) return the following informative error message:

{
"result": False,
"catch_exceptions": True,
"exception_traceback": "..."
}

catch_exceptions is on by default in command-line validation mode, and off by default in exploration mode.

meta

All Expectations accept an optional meta parameter. If meta is a valid JSON-serializable dictionary, it will be \ passed through to the expectation_result object without modification. The meta parameter can be used to add \ helpful markdown annotations to Expectations (shown below). These Expectation "notes" are rendered within \ Expectation Suite pages in Data Docs.

my_df.expect_column_values_to_be_in_set(
"my_column",
["a", "b", "c"],
meta={
"notes": {
"format": "markdown",
"content": [
"#### These are expectation notes \n - you can use markdown \n - or just strings"
]
}
}
)
# This returns:
{
"success": False,
"meta": {
"notes": {
"format": "markdown",
"content": [
"#### These are expectation notes \n - you can use markdown \n - or just strings"
]
}
}
}

mostly

mostly is a special argument that is automatically available in all column_map_expectations. mostly must be a float between 0 and 1. Great Expectations evaluates it as a percentage, allowing some wiggle room when evaluating Expectations: as long as mostly percent of rows evaluate to True, the Expectation returns "success": True.

[0,1,2,3,4,5,6,7,8,9]

my_df.expect_column_values_to_be_between(
"my_column",
min_value=0,
max_value=7
)
# This returns:
{
"success": False,
...
}

my_df.expect_column_values_to_be_between(
"my_column",
min_value=0,
max_value=7,
mostly=0.7
)
# This returns:
{
"success": True,
...
}

Expectations with mostly return exception lists even if they succeed:

my_df.expect_column_values_to_be_between(
"my_column",
min_value=0,
max_value=7,
mostly=0.7
)
# This returns:
{
"success": true
"result": {
"unexpected_percent": 0.2,
"partial_unexpected_index_list": [
8,
9
],
"partial_unexpected_list": [
8,
9
],
"unexpected_percent_nonmissing": 0.2,
"unexpected_count": 2
}
}

Dataset defaults

This default behavior for result_format, include_config, catch_exceptions can be overridden at the Dataset level:

my_dataset.set_default_expectation_argument("result_format", "SUMMARY")

In validation mode, they can be overridden using flags:

great_expectations validation csv my_dataset.csv my_expectations.json \ 
--result_format=BOOLEAN_ONLY --catch_exceptions=False --include_config=True