def _validate_semantic_types_dict(self, dataset):
"""
Validates a semantic_types dict to ensure correct formatting, that all semantic_types are recognized, and that
the semantic_types align with the column data types
Args:
dataset: A GE dataset
config: A config dictionary
Returns:
The validated semantic_types dictionary
"""
if not isinstance(self.semantic_types_dict, dict):
raise ValueError(
f"The semantic_types dict in the config must be a dictionary, but is currently a "
f"{type(self.semantic_types_dict)}. Please reformat."
)
for k, v in self.semantic_types_dict.items():
assert isinstance(v, list), (
"Entries in semantic type dict must be lists of column names e.g. "
"{'semantic_types': {'numeric': ['number_of_transactions']}}"
)
if k.upper() not in profiler_semantic_types:
raise ValueError(
f"{k} is not a recognized semantic_type. Please only include one of "
f"{profiler_semantic_types}"
)
selected_columns = [
column
for column_list in self.semantic_types_dict.values()
for column in column_list
]
if selected_columns:
for column in selected_columns:
if column not in dataset.get_table_columns():
raise ProfilerError(f"Column {column} does not exist.")
elif column in self.ignored_columns:
raise ValueError(
f"Column {column} is specified in both the semantic_types_dict and the list of "
f"ignored columns. Please remove one of these entries to proceed."
)
for semantic_type, column_list in self.semantic_types_dict.items():
for column_name in column_list:
processed_column = self.column_info.get(column_name)
if semantic_type == "datetime":
assert processed_column.get("type") in ("DATETIME", "STRING",), (
f"Column {column_name} must be a datetime column or a string but appears to be "
f"{processed_column.get('type')}"
)
elif semantic_type == "numeric":
assert processed_column.get("type") in (
"INT",
"FLOAT",
"NUMERIC",
), f"Column {column_name} must be an int or a float but appears to be {processed_column.get('type')}"
elif semantic_type in ("STRING", "VALUE_SET"):
pass
return self.semantic_types_dict
def _check_that_columns_exist(dataset, columns) -> None:
if columns:
for column in columns:
if column not in dataset.get_table_columns():
raise ProfilerError(f"Column {column} does not exist.")
def _check_that_expectations_are_available(dataset, expectations) -> None:
if expectations:
for expectation in expectations:
if expectation not in dataset.list_available_expectation_types():
raise ProfilerError(
f"Expectation {expectation} is not available.")
def _profile(cls, dataset, configuration=None):
logger.debug(f"Running profiler with configuration: {configuration}")
if configuration == "demo":
return cls._demo_profile(dataset)
existing_columns = dataset.get_table_columns()
selected_columns = existing_columns
included_expectations = []
excluded_expectations = []
if configuration:
if ("included_expectations" in configuration
and "excluded_expectations" in configuration):
raise ProfilerError(
"Please specify either `included_expectations` or `excluded_expectations`."
)
if "included_expectations" in configuration:
included_expectations = configuration["included_expectations"]
if included_expectations in [False, None, []]:
included_expectations = None
_check_that_expectations_are_available(dataset,
included_expectations)
if "excluded_expectations" in configuration:
excluded_expectations = configuration["excluded_expectations"]
if excluded_expectations in [False, None, []]:
excluded_expectations = None
_check_that_expectations_are_available(dataset,
excluded_expectations)
if ("included_columns" in configuration
and "excluded_columns" in configuration):
raise ProfilerError(
"Please specify either `excluded_columns` or `included_columns`."
)
elif "included_columns" in configuration:
selected_columns = configuration["included_columns"]
if selected_columns in [False, None, []]:
selected_columns = []
elif "excluded_columns" in configuration:
excluded_columns = configuration["excluded_columns"]
if excluded_columns in [False, None, []]:
excluded_columns = []
selected_columns = set(existing_columns) - set(
excluded_columns)
_check_that_columns_exist(dataset, selected_columns)
if included_expectations is None:
suite = cls._build_column_description_metadata(dataset)
# remove column exist expectations
suite.expectations = []
return suite
dataset.set_default_expectation_argument("catch_exceptions", False)
dataset = cls._build_table_row_count_expectation(
dataset,
excluded_expectations=excluded_expectations,
included_expectations=included_expectations,
)
dataset.set_config_value("interactive_evaluation", True)
dataset = cls._build_table_column_expectations(
dataset,
excluded_expectations=excluded_expectations,
included_expectations=included_expectations,
)
column_cache = {}
if selected_columns:
with tqdm(total=len(selected_columns),
desc="Profiling Columns",
delay=5) as pbar:
for column in selected_columns:
pbar.set_postfix_str(column)
cardinality = cls._get_column_cardinality_with_caching(
dataset, column, column_cache)
column_type = cls._get_column_type_with_caching(
dataset, column, column_cache)
if cardinality in [
ProfilerCardinality.TWO,
ProfilerCardinality.VERY_FEW,
ProfilerCardinality.FEW,
]:
cls._create_expectations_for_low_card_column(
dataset, column, column_cache)
elif cardinality in [
ProfilerCardinality.MANY,
ProfilerCardinality.VERY_MANY,
ProfilerCardinality.UNIQUE,
]:
# TODO we will want to finesse the number and types of
# expectations created here. The simple version is deny/allow list
# and the more complex version is desired per column type and
# cardinality. This deserves more thought on configuration.
dataset.expect_column_values_to_be_unique(column)
if column_type in [
ProfilerDataType.INT,
ProfilerDataType.FLOAT,
]:
cls._create_expectations_for_numeric_column(
dataset, column)
elif column_type in [ProfilerDataType.DATETIME]:
cls._create_expectations_for_datetime_column(
dataset,
column,
excluded_expectations=excluded_expectations,
included_expectations=included_expectations,
)
elif column_type in [ProfilerDataType.STRING]:
cls._create_expectations_for_string_column(
dataset,
column,
excluded_expectations=excluded_expectations,
included_expectations=included_expectations,
)
elif column_type in [ProfilerDataType.UNKNOWN]:
logger.debug(
f"Skipping expectation creation for column {column} of unknown type: {column_type}"
)
pbar.update()
if excluded_expectations:
# NOTE: we reach into a private member here because of an expected future
# refactor that will make the suite directly accessible
dataset._expectation_suite.remove_all_expectations_of_type(
excluded_expectations)
if included_expectations:
for expectation in dataset.get_expectation_suite(
discard_failed_expectations=False,
suppress_logging=True,
).expectations:
if expectation.expectation_type not in included_expectations:
try:
dataset.remove_expectation(
ExpectationConfiguration(
expectation_type=expectation.expectation_type,
kwargs=expectation.kwargs,
),
match_type="domain",
remove_multiple_matches=True,
)
except ValueError:
logger.debug(
f"Attempted to remove {expectation}, which was not found."
)
expectation_suite = cls._build_column_description_metadata(dataset)
return expectation_suite
def _profile(cls, dataset, configuration=None):
logger.debug(f"Running profiler with configuration: {configuration}")
if configuration == "demo":
return cls._demo_profile(dataset)
existing_columns = dataset.get_table_columns()
selected_columns = existing_columns
included_expectations = []
excluded_expectations = []
if configuration:
if ("included_expectations" in configuration
and "excluded_expectations" in configuration):
raise ProfilerError(
"Please specify either `included_expectations` or `excluded_expectations`."
)
if "included_expectations" in configuration:
included_expectations = configuration["included_expectations"]
if included_expectations in [False, None, []]:
included_expectations = None
_check_that_expectations_are_available(dataset,
included_expectations)
if "excluded_expectations" in configuration:
excluded_expectations = configuration["excluded_expectations"]
_check_that_expectations_are_available(dataset,
excluded_expectations)
if ("included_columns" in configuration
and "excluded_columns" in configuration):
raise ProfilerError(
"Please specify either `excluded_columns` or `included_columns`."
)
elif "included_columns" in configuration:
selected_columns = configuration["included_columns"]
elif "excluded_columns" in configuration:
excluded_columns = configuration["excluded_columns"]
if excluded_columns in [False, None, []]:
excluded_columns = []
selected_columns = set(existing_columns) - set(
excluded_columns)
_check_that_columns_exist(dataset, selected_columns)
if included_expectations is None:
suite = cls._build_column_description_metadata(dataset)
# remove column exist expectations
suite.expectations = []
return suite
dataset.set_default_expectation_argument("catch_exceptions", False)
dataset = cls._build_table_row_count_expectation(dataset,
tolerance=0.1)
dataset.set_config_value("interactive_evaluation", True)
dataset = cls._build_table_column_expectations(dataset)
column_cache = {}
if selected_columns:
for column in selected_columns:
cardinality = cls._get_column_cardinality_with_caching(
dataset, column, column_cache)
column_type = cls._get_column_type_with_caching(
dataset, column, column_cache)
if cardinality in [
ProfilerCardinality.TWO,
ProfilerCardinality.VERY_FEW,
ProfilerCardinality.FEW,
]:
cls._create_expectations_for_low_card_column(
dataset, column, column_cache)
elif cardinality in [
ProfilerCardinality.MANY,
ProfilerCardinality.VERY_MANY,
ProfilerCardinality.UNIQUE,
]:
# TODO we will want to finesse the number and types of
# expectations created here. The simple version is blacklisting
# and the more complex version is desired per column type and
# cardinality. This deserves more thought on configuration.
dataset.expect_column_values_to_be_unique(column)
if column_type in [
ProfilerDataType.INT, ProfilerDataType.FLOAT
]:
cls._create_expectations_for_numeric_column(
dataset, column)
elif column_type in [ProfilerDataType.DATETIME]:
cls._create_expectations_for_datetime_column(
dataset, column)
elif column_type in [ProfilerDataType.STRING]:
cls._create_expectations_for_string_column(
dataset, column)
elif column_type in [ProfilerDataType.UNKNOWN]:
logger.debug(
f"Skipping expectation creation for column {column} of unknown type: {column_type}"
)
if excluded_expectations:
dataset = _remove_table_expectations(dataset,
excluded_expectations)
dataset = _remove_column_expectations(dataset,
excluded_expectations)
if included_expectations:
for expectation in dataset.get_expectation_suite().expectations:
if expectation.expectation_type not in included_expectations:
try:
dataset.remove_expectation(
expectation_type=expectation.expectation_type,
expectation_kwargs=expectation.kwargs,
column=expectation.kwargs.get("column", None),
remove_multiple_matches=True,
)
except ValueError:
logger.debug(
f"Attempted to remove {expectation}, which was not found."
)
expectation_suite = cls._build_column_description_metadata(dataset)
return expectation_suite