def test_ingest_multi_process(
feature_store_session,
role,
feature_group_name,
offline_store_s3_uri,
pandas_data_frame,
):
feature_group = FeatureGroup(name=feature_group_name,
sagemaker_session=feature_store_session)
feature_group.load_feature_definitions(data_frame=pandas_data_frame)
with cleanup_feature_group(feature_group):
output = feature_group.create(
s3_uri=offline_store_s3_uri,
record_identifier_name="feature1",
event_time_feature_name="feature3",
role_arn=role,
enable_online_store=True,
)
_wait_for_feature_group_create(feature_group)
feature_group.ingest(data_frame=pandas_data_frame,
max_workers=3,
max_processes=2,
wait=True)
assert output["FeatureGroupArn"].endswith(
f"feature-group/{feature_group_name}")
def test_ingest_with_profile_name(ingestion_manager_init,
sagemaker_session_mock,
fs_runtime_client_config_mock):
sagemaker_session_mock.sagemaker_featurestore_runtime_client.meta.config = (
fs_runtime_client_config_mock)
feature_group = FeatureGroup(name="MyGroup",
sagemaker_session=sagemaker_session_mock)
df = pd.DataFrame(
dict((f"float{i}", pd.Series([2.0], dtype="float64"))
for i in range(300)))
mock_ingestion_manager_instance = Mock()
ingestion_manager_init.return_value = mock_ingestion_manager_instance
feature_group.ingest(data_frame=df,
max_workers=10,
profile_name="profile_name")
ingestion_manager_init.assert_called_once_with(
feature_group_name="MyGroup",
sagemaker_fs_runtime_client_config=fs_runtime_client_config_mock,
max_workers=10,
max_processes=1,
profile_name="profile_name",
)
mock_ingestion_manager_instance.run.assert_called_once_with(data_frame=df,
wait=True,
timeout=None)
def test_ingest_zero_workers():
feature_group = FeatureGroup(name="MyGroup", sagemaker_session=sagemaker_session_mock)
df = Mock()
with pytest.raises(RuntimeError) as error:
feature_group.ingest(data_frame=df, max_workers=0, max_processes=1)
assert "max_workers must be greater than 0." in str(error)
def save_to_feature_store():
logger.info("Save to FeatureStore started")
global feature_group
df_data = pd.read_csv(feature_s3_url)
logger.info("Read data from S3: %s", df_data.head())
feature_store_session = Session(
boto_session=boto_session,
sagemaker_client=sagemaker_client,
sagemaker_featurestore_runtime_client=featurestore_runtime)
# You can modify the following to use a bucket of your choosing
logger.info("Default bucket: %s", default_bucket)
# record identifier and event time feature names
record_identifier_feature_name = "IDpol"
event_time_feature_name = "EventTime"
current_time_sec = int(round(time.time()))
# cast object dtype to string. The SageMaker FeatureStore Python SDK will then map the string dtype to String feature type.
cast_object_to_string(df_data)
df_data[event_time_feature_name] = pd.Series([current_time_sec] *
len(df_data),
dtype="float64")
feature_group_name = 'insurance-policy-feature-group-' + strftime(
'%d-%H-%M-%S', gmtime())
logger.info("Feature Group Name: %s", feature_group_name)
# Check if feature group already exists. Create a feature group if doesn't exist.
if feature_group_exist(feature_group_name) == False:
logger.info("Feature Group: %s doesn't exist. Create a new one.",
feature_group)
feature_group = FeatureGroup(name=feature_group_name,
sagemaker_session=feature_store_session)
# append EventTime feature
# load feature definitions to the feature group. SageMaker FeatureStore Python SDK will auto-detect the data schema based on input data.
feature_group.load_feature_definitions(data_frame=df_data)
# output is suppressed
feature_group.create(
s3_uri=f"s3://{default_bucket}/{prefix}",
record_identifier_name=record_identifier_feature_name,
event_time_feature_name=event_time_feature_name,
role_arn=get_execution_role(),
enable_online_store=True)
wait_for_feature_group_creation_complete(feature_group=feature_group)
feature_group.describe()
else:
logger.info("Feature Group: %s exits", feature_group)
# Init feature group object if already exists
feature_group = FeatureGroup(name=feature_group_name,
sagemaker_session=feature_store_session)
# ingest data into feature store
feature_group.ingest(data_frame=df_data, max_workers=5, wait=True)
def test_ingest_default_max_workers(ingestion_manager_init, sagemaker_session_mock):
feature_group = FeatureGroup(name="MyGroup", sagemaker_session=sagemaker_session_mock)
df = pd.DataFrame({"float": pd.Series([2.0], dtype="float64")})
mock_ingestion_manager_instance = Mock()
ingestion_manager_init.return_value = mock_ingestion_manager_instance
feature_group.ingest(data_frame=df)
ingestion_manager_init.assert_called_once_with(
feature_group_name="MyGroup",
sagemaker_session=sagemaker_session_mock,
data_frame=df,
max_workers=1,
)
mock_ingestion_manager_instance.run.assert_called_once_with(wait=True, timeout=None)
def test_ingest(ingestion_manager_init, sagemaker_session_mock):
feature_group = FeatureGroup(name="MyGroup", sagemaker_session=sagemaker_session_mock)
df = pd.DataFrame(dict((f"float{i}", pd.Series([2.0], dtype="float64")) for i in range(300)))
mock_ingestion_manager_instance = Mock()
ingestion_manager_init.return_value = mock_ingestion_manager_instance
feature_group.ingest(data_frame=df, max_workers=10)
ingestion_manager_init.assert_called_once_with(
feature_group_name="MyGroup",
sagemaker_session=sagemaker_session_mock,
data_frame=df,
max_workers=10,
)
mock_ingestion_manager_instance.run.assert_called_once_with(wait=True, timeout=None)
def test_ingest_without_string_feature(
feature_store_session,
role,
feature_group_name,
offline_store_s3_uri,
pandas_data_frame_without_string,
):
feature_group = FeatureGroup(name=feature_group_name,
sagemaker_session=feature_store_session)
feature_group.load_feature_definitions(
data_frame=pandas_data_frame_without_string)
with cleanup_feature_group(feature_group):
output = feature_group.create(
s3_uri=offline_store_s3_uri,
record_identifier_name="feature1",
event_time_feature_name="feature2",
role_arn=role,
enable_online_store=True,
)
_wait_for_feature_group_create(feature_group)
ingestion_manager = feature_group.ingest(
data_frame=pandas_data_frame_without_string,
max_workers=3,
wait=False)
ingestion_manager.wait()
assert output["FeatureGroupArn"].endswith(
f"feature-group/{feature_group_name}")
def test_create_feature_store(
feature_store_session,
role,
feature_group_name,
offline_store_s3_uri,
pandas_data_frame,
record,
create_table_ddl,
):
feature_group = FeatureGroup(name=feature_group_name,
sagemaker_session=feature_store_session)
feature_group.load_feature_definitions(data_frame=pandas_data_frame)
with cleanup_feature_group(feature_group):
output = feature_group.create(
s3_uri=offline_store_s3_uri,
record_identifier_name="feature1",
event_time_feature_name="feature3",
role_arn=role,
enable_online_store=True,
)
_wait_for_feature_group_create(feature_group)
# Ingest data
feature_group.put_record(record=record)
ingestion_manager = feature_group.ingest(data_frame=pandas_data_frame,
max_workers=3,
wait=False)
ingestion_manager.wait()
assert 0 == len(ingestion_manager.failed_rows)
# Query the integrated Glue table.
athena_query = feature_group.athena_query()
df = DataFrame()
with timeout(minutes=10):
while df.shape[0] < 11:
athena_query.run(
query_string=f'SELECT * FROM "{athena_query.table_name}"',
output_location=f"{offline_store_s3_uri}/query_results",
)
athena_query.wait()
assert "SUCCEEDED" == athena_query.get_query_execution().get(
"QueryExecution").get("Status").get("State")
df = athena_query.as_dataframe()
print(f"Found {df.shape[0]} records.")
time.sleep(60)
assert df.shape[0] == 11
nans = pd.isna(df.loc[df["feature1"].isin([5, 6, 7, 8,
9])]["feature4"])
for is_na in nans.items():
assert is_na
assert (create_table_ddl.format(
feature_group_name=feature_group_name,
region=feature_store_session.boto_session.region_name,
account=feature_store_session.account_id(),
) == feature_group.as_hive_ddl())
assert output["FeatureGroupArn"].endswith(
f"feature-group/{feature_group_name}")
role_arn=role,
enable_online_store=True,
description="1.8M+ tokenized camera reviews from the Amazon Customer Reviews dataset",
tags=[
{ 'Key': 'Dataset', 'Value': 'amazon customer reviews' },
{ 'Key': 'Subset', 'Value': 'cameras' },
{ 'Key': 'Owner', 'Value': 'Julien Simon' }
]
)
# Wait for feature group to be ready
while feature_group.describe().get("FeatureGroupStatus") != 'Created':
sleep(1)
print('Feature group created')
# Ingest data
print('Ingesting data...')
try:
feature_group.ingest(data_frame=data, max_workers=max_workers, wait=True)
except Exception:
pass
print('Waiting...')
# Wait for 10 minutes to make sure data has flowed to the offline store
# https://docs.aws.amazon.com/sagemaker/latest/dg/feature-store-offline.html
sleep(600)
# Save feature group name
with open('/opt/ml/processing/output/feature_group_name.txt', 'w') as f:
f.write(fg_name)
print('Job complete')
class FeatureGroupDataSet(AbstractDataSet):
def __init__(
self,
name: str,
s3_uri: str,
record_identifier_name: str,
event_time_name: str,
query: str,
description: str = None,
):
region = boto3.Session().region_name
boto_session = boto3.Session(region_name=region)
sagemaker_client = boto_session.client(
service_name="sagemaker", region_name=region
)
featurestore_runtime = boto_session.client(
service_name="sagemaker-featurestore-runtime", region_name=region
)
feature_store_session = Session(
boto_session=boto_session,
sagemaker_client=sagemaker_client,
sagemaker_featurestore_runtime_client=featurestore_runtime,
)
iam = boto3.client("iam")
role = iam.get_role(RoleName="AmazonSageMaker-ExecutionRole")["Role"]["Arn"]
# you can also suffix the feature group name with pipeline git version
self._feature_group = FeatureGroup(
name=name, sagemaker_session=feature_store_session
)
self._description = description
self._s3_uri = s3_uri
self._role = role
self._record_identifier_name = record_identifier_name
self._event_time_name = event_time_name
self._query = query
def _wait_for_feature_group_creation_complete(self):
status = self._feature_group.describe().get("FeatureGroupStatus")
while status == "Creating":
logger.info("Waiting for Feature Group Creation")
time.sleep(5)
status = self._feature_group.describe().get("FeatureGroupStatus")
if status != "Created":
raise RuntimeError(
f"Failed to create feature group {self._feature_group.name}"
)
logger.info("FeatureGroup %s successfully created.", self._feature_group.name)
def _describe(self):
return dict(feature_group=self._feature_group)
def _save(self, data):
self._feature_group.load_feature_definitions(data)
try:
self._feature_group.create(
description=self._description,
s3_uri=self._s3_uri,
record_identifier_name=self._record_identifier_name,
event_time_feature_name=self._event_time_name,
role_arn=self._role,
enable_online_store=True,
)
self._wait_for_feature_group_creation_complete()
except Exception as exc:
if (
f"Resource Already Exists: FeatureGroup with name {self._feature_group.name} already exists"
in str(exc)
):
pass
else:
raise
self._feature_group.ingest(data[:10]) # just for demo purpose
def _load(self) -> pd.DataFrame:
query = self._feature_group.athena_query()
print(self._query.format(table_name=query.table_name))
query.run(
self._query.format(table_name=query.table_name),
output_location=f"{self._s3_uri}/query_results/",
)
query.wait()
return query.as_dataframe()
