def test_model_log():
with TempDir(chdr=True) as tmp:
experiment_id = kiwi.create_experiment("test")
sig = ModelSignature(inputs=Schema([ColSpec("integer", "x"), ColSpec("integer", "y")]),
outputs=Schema([ColSpec(name=None, type="double")]))
input_example = {"x": 1, "y": 2}
with kiwi.start_run(experiment_id=experiment_id) as r:
Model.log("some/path", TestFlavor,
signature=sig,
input_example=input_example)
local_path = _download_artifact_from_uri("runs:/{}/some/path".format(r.info.run_id),
output_path=tmp.path(""))
loaded_model = Model.load(os.path.join(local_path, "MLmodel"))
assert loaded_model.run_id == r.info.run_id
assert loaded_model.artifact_path == "some/path"
assert loaded_model.flavors == {
"flavor1": {"a": 1, "b": 2},
"flavor2": {"x": 1, "y": 2},
}
assert loaded_model.signature == sig
path = os.path.join(local_path, loaded_model.saved_input_example_info["artifact_path"])
x = _dataframe_from_json(path)
assert x.to_dict(orient="records")[0] == input_example
def test_model_log_load_no_active_run(sklearn_knn_model, iris_data, tmpdir):
sk_model_path = os.path.join(str(tmpdir), "knn.pkl")
with open(sk_model_path, "wb") as f:
pickle.dump(sklearn_knn_model, f)
pyfunc_artifact_path = "pyfunc_model"
assert kiwi.active_run() is None
kiwi.pyfunc.log_model(artifact_path=pyfunc_artifact_path,
data_path=sk_model_path,
loader_module=os.path.basename(__file__)[:-3],
code_path=[__file__])
pyfunc_model_path = _download_artifact_from_uri(
"runs:/{run_id}/{artifact_path}".format(
run_id=kiwi.active_run().info.run_id,
artifact_path=pyfunc_artifact_path))
model_config = Model.load(os.path.join(pyfunc_model_path, "MLmodel"))
assert kiwi.pyfunc.FLAVOR_NAME in model_config.flavors
assert kiwi.pyfunc.PY_VERSION in model_config.flavors[
kiwi.pyfunc.FLAVOR_NAME]
reloaded_model = kiwi.pyfunc.load_pyfunc(pyfunc_model_path)
np.testing.assert_array_equal(sklearn_knn_model.predict(iris_data[0]),
reloaded_model.predict(iris_data[0]))
kiwi.end_run()
def test_load_model_with_missing_cloudpickle_version_logs_warning(model_path):
class TestModel(kiwi.pyfunc.PythonModel):
def predict(self, context, model_input):
return model_input
kiwi.pyfunc.save_model(path=model_path, python_model=TestModel())
model_config_path = os.path.join(model_path, "MLmodel")
model_config = Model.load(model_config_path)
del model_config.flavors[kiwi.pyfunc.FLAVOR_NAME][
kiwi.pyfunc.model.CONFIG_KEY_CLOUDPICKLE_VERSION]
model_config.save(model_config_path)
log_messages = []
def custom_warn(message_text, *args, **kwargs):
log_messages.append(message_text % args % kwargs)
with mock.patch("mlflow.pyfunc._logger.warning") as warn_mock:
warn_mock.side_effect = custom_warn
kiwi.pyfunc.load_pyfunc(model_uri=model_path)
assert any([(
"The version of CloudPickle used to save the model could not be found in the MLmodel"
" configuration") in log_message for log_message in log_messages])
def deploy(app_name,
model_uri,
execution_role_arn=None,
bucket=None,
image_url=None,
region_name="us-west-2",
mode=DEPLOYMENT_MODE_CREATE,
archive=False,
instance_type=DEFAULT_SAGEMAKER_INSTANCE_TYPE,
instance_count=DEFAULT_SAGEMAKER_INSTANCE_COUNT,
vpc_config=None,
flavor=None,
synchronous=True,
timeout_seconds=1200):
"""
Deploy an MLflow model on AWS SageMaker.
The currently active AWS account must have correct permissions set up.
This function creates a SageMaker endpoint. For more information about the input data
formats accepted by this endpoint, see the
:ref:`MLflow deployment tools documentation <sagemaker_deployment>`.
:param app_name: Name of the deployed application.
:param model_uri: The location, in URI format, of the MLflow model to deploy to SageMaker.
For example:
- ``/Users/me/path/to/local/model``
- ``relative/path/to/local/model``
- ``s3://my_bucket/path/to/model``
- ``runs:/<mlflow_run_id>/run-relative/path/to/model``
- ``models:/<model_name>/<model_version>``
- ``models:/<model_name>/<stage>``
For more information about supported URI schemes, see
`Referencing Artifacts <https://www.mlflow.org/docs/latest/concepts.html#
artifact-locations>`_.
:param execution_role_arn: The name of an IAM role granting the SageMaker service permissions to
access the specified Docker image and S3 bucket containing MLflow
model artifacts. If unspecified, the currently-assumed role will be
used. This execution role is passed to the SageMaker service when
creating a SageMaker model from the specified MLflow model. It is
passed as the ``ExecutionRoleArn`` parameter of the `SageMaker
CreateModel API call <https://docs.aws.amazon.com/sagemaker/latest/
dg/API_CreateModel.html>`_. This role is *not* assumed for any other
call. For more information about SageMaker execution roles for model
creation, see
https://docs.aws.amazon.com/sagemaker/latest/dg/sagemaker-roles.html.
:param bucket: S3 bucket where model artifacts will be stored. Defaults to a
SageMaker-compatible bucket name.
:param image_url: URL of the ECR-hosted Docker image the model should be deployed into, produced
by ``mlflow sagemaker build-and-push-container``. This parameter can also
be specified by the environment variable ``MLFLOW_SAGEMAKER_DEPLOY_IMG_URL``.
:param region_name: Name of the AWS region to which to deploy the application.
:param mode: The mode in which to deploy the application. Must be one of the following:
``mlflow.sagemaker.DEPLOYMENT_MODE_CREATE``
Create an application with the specified name and model. This fails if an
application of the same name already exists.
``mlflow.sagemaker.DEPLOYMENT_MODE_REPLACE``
If an application of the specified name exists, its model(s) is replaced with
the specified model. If no such application exists, it is created with the
specified name and model.
``mlflow.sagemaker.DEPLOYMENT_MODE_ADD``
Add the specified model to a pre-existing application with the specified name,
if one exists. If the application does not exist, a new application is created
with the specified name and model. NOTE: If the application **already exists**,
the specified model is added to the application's corresponding SageMaker
endpoint with an initial weight of zero (0). To route traffic to the model,
update the application's associated endpoint configuration using either the
AWS console or the ``UpdateEndpointWeightsAndCapacities`` function defined in
https://docs.aws.amazon.com/sagemaker/latest/dg/API_UpdateEndpointWeightsAndCapacities.html.
:param archive: If ``True``, any pre-existing SageMaker application resources that become
inactive (i.e. as a result of deploying in
``mlflow.sagemaker.DEPLOYMENT_MODE_REPLACE`` mode) are preserved.
These resources may include unused SageMaker models and endpoint configurations
that were associated with a prior version of the application endpoint. If
``False``, these resources are deleted. In order to use ``archive=False``,
``deploy()`` must be executed synchronously with ``synchronous=True``.
:param instance_type: The type of SageMaker ML instance on which to deploy the model. For a list
of supported instance types, see
https://aws.amazon.com/sagemaker/pricing/instance-types/.
:param instance_count: The number of SageMaker ML instances on which to deploy the model.
:param vpc_config: A dictionary specifying the VPC configuration to use when creating the
new SageMaker model associated with this application. The acceptable values
for this parameter are identical to those of the ``VpcConfig`` parameter in
the `SageMaker boto3 client's create_model method
<https://boto3.readthedocs.io/en/latest/reference/services/sagemaker.html
#SageMaker.Client.create_model>`_. For more information, see
https://docs.aws.amazon.com/sagemaker/latest/dg/API_VpcConfig.html.
.. code-block:: python
:caption: Example
import mlflow.sagemaker as mfs
vpc_config = {
'SecurityGroupIds': [
'sg-123456abc',
],
'Subnets': [
'subnet-123456abc',
]
}
mfs.deploy(..., vpc_config=vpc_config)
:param flavor: The name of the flavor of the model to use for deployment. Must be either
``None`` or one of mlflow.sagemaker.SUPPORTED_DEPLOYMENT_FLAVORS. If ``None``,
a flavor is automatically selected from the model's available flavors. If the
specified flavor is not present or not supported for deployment, an exception
will be thrown.
:param synchronous: If ``True``, this function will block until the deployment process succeeds
or encounters an irrecoverable failure. If ``False``, this function will
return immediately after starting the deployment process. It will not wait
for the deployment process to complete; in this case, the caller is
responsible for monitoring the health and status of the pending deployment
via native SageMaker APIs or the AWS console.
:param timeout_seconds: If ``synchronous`` is ``True``, the deployment process will return after
the specified number of seconds if no definitive result (success or
failure) is achieved. Once the function returns, the caller is
responsible for monitoring the health and status of the pending
deployment using native SageMaker APIs or the AWS console. If
``synchronous`` is ``False``, this parameter is ignored.
"""
import boto3
if (not archive) and (not synchronous):
raise MlflowException(message=(
"Resources must be archived when `deploy()` is executed in non-synchronous mode."
" Either set `synchronous=True` or `archive=True`."),
error_code=INVALID_PARAMETER_VALUE)
if mode not in DEPLOYMENT_MODES:
raise MlflowException(
message="`mode` must be one of: {deployment_modes}".format(
deployment_modes=",".join(DEPLOYMENT_MODES)),
error_code=INVALID_PARAMETER_VALUE)
model_path = _download_artifact_from_uri(model_uri)
model_config_path = os.path.join(model_path, MLMODEL_FILE_NAME)
if not os.path.exists(model_config_path):
raise MlflowException(message=(
"Failed to find {} configuration within the specified model's"
" root directory.").format(MLMODEL_FILE_NAME),
error_code=INVALID_PARAMETER_VALUE)
model_config = Model.load(model_config_path)
if flavor is None:
flavor = _get_preferred_deployment_flavor(model_config)
else:
_validate_deployment_flavor(model_config, flavor)
_logger.info("Using the %s flavor for deployment!", flavor)
sage_client = boto3.client('sagemaker', region_name=region_name)
s3_client = boto3.client('s3', region_name=region_name)
endpoint_exists = _find_endpoint(endpoint_name=app_name,
sage_client=sage_client) is not None
if endpoint_exists and mode == DEPLOYMENT_MODE_CREATE:
raise MlflowException(message=(
"You are attempting to deploy an application with name: {application_name} in"
" '{mode_create}' mode. However, an application with the same name already"
" exists. If you want to update this application, deploy in '{mode_add}' or"
" '{mode_replace}' mode.".format(
application_name=app_name,
mode_create=DEPLOYMENT_MODE_CREATE,
mode_add=DEPLOYMENT_MODE_ADD,
mode_replace=DEPLOYMENT_MODE_REPLACE)),
error_code=INVALID_PARAMETER_VALUE)
model_name = _get_sagemaker_model_name(endpoint_name=app_name)
if not image_url:
image_url = _get_default_image_url(region_name=region_name)
if not execution_role_arn:
execution_role_arn = _get_assumed_role_arn()
if not bucket:
_logger.info(
"No model data bucket specified, using the default bucket")
bucket = _get_default_s3_bucket(region_name)
model_s3_path = _upload_s3(local_model_path=model_path,
bucket=bucket,
prefix=model_name,
region_name=region_name,
s3_client=s3_client)
if endpoint_exists:
deployment_operation = _update_sagemaker_endpoint(
endpoint_name=app_name,
model_name=model_name,
model_s3_path=model_s3_path,
model_uri=model_uri,
image_url=image_url,
flavor=flavor,
instance_type=instance_type,
instance_count=instance_count,
vpc_config=vpc_config,
mode=mode,
role=execution_role_arn,
sage_client=sage_client,
s3_client=s3_client)
else:
deployment_operation = _create_sagemaker_endpoint(
endpoint_name=app_name,
model_name=model_name,
model_s3_path=model_s3_path,
model_uri=model_uri,
image_url=image_url,
flavor=flavor,
instance_type=instance_type,
instance_count=instance_count,
vpc_config=vpc_config,
role=execution_role_arn,
sage_client=sage_client)
if synchronous:
_logger.info("Waiting for the deployment operation to complete...")
operation_status = deployment_operation.await_completion(
timeout_seconds=timeout_seconds)
if operation_status.state == _SageMakerOperationStatus.STATE_SUCCEEDED:
_logger.info(
"The deployment operation completed successfully with message: \"%s\"",
operation_status.message)
else:
raise MlflowException(
"The deployment operation failed with the following error message:"
" \"{error_message}\"".format(
error_message=operation_status.message))
if not archive:
deployment_operation.clean_up()
