expect_column_mean_to_be_between expectation unsuccessful for empty DataFrame

[mlondschien]

Describe the bug The expect_column_mean_to_be_between expectations fails if used together with a row_condition that evaluates False for all rows. This is inconsistent with other expectations.

To Reproduce:

In [1]: import great_expectations as ge
   ...: import pandas as pd
   ...: 
   ...: df = pd.DataFrame({"i": [0, 1, 0, 1], "y": ["a", "a", "a", "a"]})
   ...: 
   ...: expectations = [
   ...:     {
   ...:         "expectation_type": "expect_column_values_to_not_be_null",
   ...:         "kwargs": {
   ...:             "column": "i",
   ...:             "row_condition": "y == 'b'",
   ...:             "condition_parser": "pandas",
   ...:         },
   ...:     },
   ...:     {
   ...:         "expectation_type": "expect_column_values_to_be_null",
   ...:         "kwargs": {
   ...:             "column": "i",
   ...:             "row_condition": "y == 'b'",
   ...:             "condition_parser": "pandas",
   ...:         },
   ...:     },
   ...:     {
   ...:         "expectation_type": "expect_column_values_to_be_in_set",
   ...:         "kwargs": {
   ...:             "column": "i",
   ...:             "value_set": [0, 1],
   ...:             "row_condition": "y == 'b'",
   ...:             "condition_parser": "pandas",
   ...:         },
   ...:     },
   ...:     {
   ...:         "expectation_type": "expect_column_mean_to_be_between",
   ...:         "kwargs": {
   ...:             "column": "i",
   ...:             "min_value": 0,
   ...:             "max_value": 1,
   ...:             "row_condition": "y == 'b'",
   ...:             "condition_parser": "pandas",
   ...:         },
   ...:     },
   ...: ]
   ...: 
   ...: expectation_suite = ge.core.ExpectationSuite(
   ...:     "expectation_suite",
   ...:     expectations=[ge.core.ExpectationConfiguration(**e) for e in expectations],
   ...: )
   ...: validation_results = ge.from_pandas(df).validate(expectation_suite)
   ...: 
   ...: [result.success for result in validation_results.results]
Out[1]: [True, True, True, False]

Expected Behaviour: The expect_column_mean_to_be_between expectation should succeed like other expectations.

Environment (please complete the following information):

 ~ $ conda list great-expectations
# packages in environment at /home/mlondschien/anaconda3/envs/quantcore.thek:
#
# Name                    Version                   Build  Channel
great-expectations        0.13.19            pyha770c72_0    conda-forge

Additional context We apply the same expectation suite on different subsets of a large table. Some expectations do not make sense for a specific subset, so we use the row_condition to filter these.

Details:

In [2]: validation_results Out[2]: { "statistics": { "evaluated_expectations": 4, "successful_expectations": 3, "unsuccessful_expectations": 1, "success_percent": 75.0 }, "results": [ { "expectation_config": { "expectation_type": "expect_column_values_to_not_be_null", "kwargs": { "column": "i", "row_condition": "y == 'b'", "condition_parser": "pandas" }, "meta": {} }, "result": { "element_count": 0, "unexpected_count": 0, "unexpected_percent": null, "partial_unexpected_list": [] }, "success": true, "meta": {}, "exception_info": { "raised_exception": false, "exception_message": null, "exception_traceback": null } }, { "expectation_config": { "expectation_type": "expect_column_values_to_be_null", "kwargs": { "column": "i", "row_condition": "y == 'b'", "condition_parser": "pandas" }, "meta": {} }, "result": { "element_count": 0, "unexpected_count": 0, "unexpected_percent": null, "partial_unexpected_list": [] }, "success": true, "meta": {}, "exception_info": { "raised_exception": false, "exception_message": null, "exception_traceback": null } }, { "expectation_config": { "expectation_type": "expect_column_values_to_be_in_set", "kwargs": { "column": "i", "value_set": [ 0, 1 ], "row_condition": "y == 'b'", "condition_parser": "pandas" }, "meta": {} }, "result": { "element_count": 0, "missing_count": 0, "missing_percent": null, "unexpected_count": 0, "unexpected_percent": null, "unexpected_percent_nonmissing": null, "partial_unexpected_list": [] }, "success": true, "meta": {}, "exception_info": { "raised_exception": false, "exception_message": null, "exception_traceback": null } }, { "expectation_config": { "expectation_type": "expect_column_mean_to_be_between", "kwargs": { "column": "i", "min_value": 0, "max_value": 1, "row_condition": "y == 'b'", "condition_parser": "pandas" }, "meta": {} }, "result": { "observed_value": null, "element_count": 0, "missing_count": null, "missing_percent": null }, "success": false, "meta": {}, "exception_info": { "raised_exception": false, "exception_message": null, "exception_traceback": null } } ], "evaluation_parameters": {}, "success": false, "meta": { "great_expectations_version": "0.13.2", "expectation_suite_name": "expectation_suite", "run_id": { "run_time": "2021-04-26T08:14:57.037509+00:00", "run_name": null }, "batch_kwargs": { "ge_batch_id": "7cc86ff4-a667-11eb-9e90-482ae30df8e3" }, "batch_markers": {}, "batch_parameters": {}, "validation_time": "20210426T081457.037357Z" } }

[eugmandel]

@mlondschien Thank you for reporting this!

[shearer12345]

I hit this for expect_column_values_to_be_in_type_list.py also, so wonder if this may exist for many expectations?

Would it make sense for expectations on column values (or aggregates thereof) to exit early with success on empty dataframes?

[austiezr]

Hey @mlondschien ! After significant discussion on our philosophy in this are, we believe the behavior you're seeing to be the correct one, namely that:

• Aggregate Expectations on Empty Dataframes → Fail

• Aggregate Expectations with Row Conditions that return Empty Dataframes → Fail

• Map Expectations on Empty Dataframes → If there are no rows, the expectation should pass

• Map Expectations with Row Conditions that Return Empty Dataframes → If there are no rows, the expectation should pass

We believe this should hold true for the majority of extant expectations, and will view behavior outside of this paradigm as unexpected.