def test_algorithm_hyperparameter_categorical_range(sagemaker_session):
hyperparameters = [
{
'Description': 'A continuous hyperparameter',
'Type': 'Categorical',
'Name': 'hp1',
'Range': {'CategoricalParameterRangeSpecification': {'Values': ['TF', 'MXNet']}},
'IsTunable': True,
'IsRequired': False,
'DefaultValue': '100',
}
]
some_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
some_algo['TrainingSpecification']['SupportedHyperParameters'] = hyperparameters
sagemaker_session.sagemaker_client.describe_algorithm = Mock(return_value=some_algo)
estimator = AlgorithmEstimator(
algorithm_arn='arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees',
role='SageMakerRole',
train_instance_type='ml.m4.2xlarge',
train_instance_count=1,
sagemaker_session=sagemaker_session,
)
estimator.set_hyperparameters(hp1='MXNet')
estimator.set_hyperparameters(hp1='TF')
with pytest.raises(ValueError):
estimator.set_hyperparameters(hp1='Chainer')
with pytest.raises(ValueError):
estimator.set_hyperparameters(hp1='MxNET')
def test_algorithm_create_transformer(create_model, sagemaker_session):
sagemaker_session.sagemaker_client.describe_algorithm = Mock(
return_value=DESCRIBE_ALGORITHM_RESPONSE)
estimator = AlgorithmEstimator(
algorithm_arn=
'arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees',
role='SageMakerRole',
train_instance_type='ml.m4.xlarge',
train_instance_count=1,
sagemaker_session=sagemaker_session,
)
estimator.latest_training_job = _TrainingJob(sagemaker_session,
'some-job-name')
model = Mock()
model.name = 'my-model'
create_model.return_value = model
transformer = estimator.transformer(instance_count=1,
instance_type='ml.m4.xlarge')
assert isinstance(transformer, Transformer)
create_model.assert_called()
assert transformer.model_name == 'my-model'
def test_algorithm_hyperparameter_categorical_range(session):
hyperparameters = [
{
"Description": "A continuous hyperparameter",
"Type": "Categorical",
"Name": "hp1",
"Range": {"CategoricalParameterRangeSpecification": {"Values": ["TF", "MXNet"]}},
"IsTunable": True,
"IsRequired": False,
"DefaultValue": "100",
}
]
some_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
some_algo["TrainingSpecification"]["SupportedHyperParameters"] = hyperparameters
session.sagemaker_client.describe_algorithm = Mock(return_value=some_algo)
estimator = AlgorithmEstimator(
algorithm_arn="arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
train_instance_type="ml.m4.2xlarge",
train_instance_count=1,
sagemaker_session=session,
)
estimator.set_hyperparameters(hp1="MXNet")
estimator.set_hyperparameters(hp1="TF")
with pytest.raises(ValueError):
estimator.set_hyperparameters(hp1="Chainer")
with pytest.raises(ValueError):
estimator.set_hyperparameters(hp1="MxNET")
def test_algorithm_create_transformer_with_product_id(create_model,
sagemaker_session):
response = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
response['ProductId'] = 'some-product-id'
sagemaker_session.sagemaker_client.describe_algorithm = Mock(
return_value=response)
estimator = AlgorithmEstimator(
algorithm_arn=
'arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees',
role='SageMakerRole',
train_instance_type='ml.m4.xlarge',
train_instance_count=1,
sagemaker_session=sagemaker_session,
)
estimator.latest_training_job = _TrainingJob(sagemaker_session,
'some-job-name')
model = Mock()
model.name = 'my-model'
create_model.return_value = model
transformer = estimator.transformer(instance_count=1,
instance_type='ml.m4.xlarge')
assert transformer.env is None
def test_algorithm_train_instance_types_valid_instance_types(session):
describe_algo_response = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
instance_types = ["ml.m4.xlarge", "ml.m5.2xlarge"]
describe_algo_response["TrainingSpecification"][
"SupportedTrainingInstanceTypes"] = instance_types
session.sagemaker_client.describe_algorithm = Mock(
return_value=describe_algo_response)
AlgorithmEstimator(
algorithm_arn=
"arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
instance_type="ml.m4.xlarge",
instance_count=1,
sagemaker_session=session,
)
AlgorithmEstimator(
algorithm_arn=
"arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
instance_type="ml.m5.2xlarge",
instance_count=1,
sagemaker_session=session,
)
def test_algorithm_train_instance_types_valid_instance_types(sagemaker_session):
describe_algo_response = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
train_instance_types = ['ml.m4.xlarge', 'ml.m5.2xlarge']
describe_algo_response['TrainingSpecification'][
'SupportedTrainingInstanceTypes'
] = train_instance_types
sagemaker_session.sagemaker_client.describe_algorithm = Mock(
return_value=describe_algo_response
)
AlgorithmEstimator(
algorithm_arn='arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees',
role='SageMakerRole',
train_instance_type='ml.m4.xlarge',
train_instance_count=1,
sagemaker_session=sagemaker_session,
)
AlgorithmEstimator(
algorithm_arn='arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees',
role='SageMakerRole',
train_instance_type='ml.m5.2xlarge',
train_instance_count=1,
sagemaker_session=sagemaker_session,
)
def test_prepare_for_training_with_name_based_on_algorithm(sagemaker_session):
estimator = AlgorithmEstimator(
algorithm_arn=
'arn:aws:sagemaker:us-west-2:1234:algorithm/scikit-decision-trees-1542410022',
role='some_image',
train_instance_count=1,
train_instance_type='ml.m4.xlarge',
sagemaker_session=sagemaker_session)
estimator._prepare_for_training()
assert 'scikit-decision-trees-1542410022' in estimator._current_job_name
def test_algorithm_enable_network_isolation_no_product_id(session):
session.sagemaker_client.describe_algorithm = Mock(return_value=DESCRIBE_ALGORITHM_RESPONSE)
estimator = AlgorithmEstimator(
algorithm_arn="arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
train_instance_type="ml.m4.xlarge",
train_instance_count=1,
sagemaker_session=session,
)
network_isolation = estimator.enable_network_isolation()
assert network_isolation is False
def test_algorithm_create_transformer_without_completed_training_job(session):
session.sagemaker_client.describe_algorithm = Mock(return_value=DESCRIBE_ALGORITHM_RESPONSE)
estimator = AlgorithmEstimator(
algorithm_arn="arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
train_instance_type="ml.m4.xlarge",
train_instance_count=1,
sagemaker_session=session,
)
with pytest.raises(RuntimeError) as error:
estimator.transformer(instance_count=1, instance_type="ml.m4.xlarge")
assert "No finished training job found associated with this estimator" in str(error)
def test_algorithm_hyperparameter_continuous_range_valid_range(session):
hyperparameters = [{
"Description": "A continuous hyperparameter",
"Type": "Continuous",
"Name": "max_leaf_nodes",
"Range": {
"ContinuousParameterRangeSpecification": {
"MinValue": "0.0",
"MaxValue": "1.0"
}
},
"IsTunable": True,
"IsRequired": False,
"DefaultValue": "100",
}]
some_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
some_algo["TrainingSpecification"][
"SupportedHyperParameters"] = hyperparameters
session.sagemaker_client.describe_algorithm = Mock(return_value=some_algo)
estimator = AlgorithmEstimator(
algorithm_arn=
"arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
train_instance_type="ml.m4.2xlarge",
train_instance_count=1,
sagemaker_session=session,
)
estimator.set_hyperparameters(max_leaf_nodes=0)
estimator.set_hyperparameters(max_leaf_nodes=1.0)
estimator.set_hyperparameters(max_leaf_nodes=0.5)
estimator.set_hyperparameters(max_leaf_nodes=1)
def test_algorithm_hyperparameter_continuous_range_valid_range(sagemaker_session):
hyperparameters = [
{
'Description': 'A continuous hyperparameter',
'Type': 'Continuous',
'Name': 'max_leaf_nodes',
'Range': {
'ContinuousParameterRangeSpecification': {'MinValue': '0.0', 'MaxValue': '1.0'}
},
'IsTunable': True,
'IsRequired': False,
'DefaultValue': '100',
}
]
some_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
some_algo['TrainingSpecification']['SupportedHyperParameters'] = hyperparameters
sagemaker_session.sagemaker_client.describe_algorithm = Mock(return_value=some_algo)
estimator = AlgorithmEstimator(
algorithm_arn='arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees',
role='SageMakerRole',
train_instance_type='ml.m4.2xlarge',
train_instance_count=1,
sagemaker_session=sagemaker_session,
)
estimator.set_hyperparameters(max_leaf_nodes=0)
estimator.set_hyperparameters(max_leaf_nodes=1.0)
estimator.set_hyperparameters(max_leaf_nodes=0.5)
estimator.set_hyperparameters(max_leaf_nodes=1)
def test_algorithm_enable_network_isolation_with_product_id(session):
response = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
response["ProductId"] = "some-product-id"
session.sagemaker_client.describe_algorithm = Mock(return_value=response)
estimator = AlgorithmEstimator(
algorithm_arn="arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
train_instance_type="ml.m4.xlarge",
train_instance_count=1,
sagemaker_session=session,
)
network_isolation = estimator.enable_network_isolation()
assert network_isolation is True
def test_algorithm_required_hyperparameters_are_provided(session):
hyperparameters = [
{
"Description": "A categorical hyperparameter",
"Type": "Categorical",
"Name": "hp1",
"Range": {
"CategoricalParameterRangeSpecification": {
"Values": ["TF", "MXNet"]
}
},
"IsTunable": True,
"IsRequired": True,
},
{
"Name": "hp2",
"Description": "A categorical hyperparameter",
"Type": "Categorical",
"IsTunable": False,
"IsRequired": True,
},
{
"Name": "free_text_hp1",
"Description": "You can write anything here",
"Type": "FreeText",
"IsTunable": False,
"IsRequired": True,
},
]
some_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
some_algo["TrainingSpecification"][
"SupportedHyperParameters"] = hyperparameters
session.sagemaker_client.describe_algorithm = Mock(return_value=some_algo)
estimator = AlgorithmEstimator(
algorithm_arn=
"arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
train_instance_type="ml.m4.2xlarge",
train_instance_count=1,
sagemaker_session=session,
)
# All 3 Hyperparameters are provided
estimator.set_hyperparameters(hp1="TF", hp2="TF2", free_text_hp1="Hello!")
def test_algorithm_distributed_training_validation(session):
distributed_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
distributed_algo["TrainingSpecification"][
"SupportsDistributedTraining"] = True
single_instance_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
single_instance_algo["TrainingSpecification"][
"SupportsDistributedTraining"] = False
session.sagemaker_client.describe_algorithm = Mock(
return_value=distributed_algo)
# Distributed training should work for Distributed and Single instance.
AlgorithmEstimator(
algorithm_arn=
"arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
train_instance_type="ml.m4.xlarge",
train_instance_count=1,
sagemaker_session=session,
)
AlgorithmEstimator(
algorithm_arn=
"arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
train_instance_type="ml.m4.xlarge",
train_instance_count=2,
sagemaker_session=session,
)
session.sagemaker_client.describe_algorithm = Mock(
return_value=single_instance_algo)
# distributed training on a single instance algorithm should fail.
with pytest.raises(ValueError):
AlgorithmEstimator(
algorithm_arn=
"arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
train_instance_type="ml.m5.2xlarge",
train_instance_count=2,
sagemaker_session=session,
)
def test_algorithm_required_hyperparameters_are_provided(sagemaker_session):
hyperparameters = [{
'Description': 'A categorical hyperparameter',
'Type': 'Categorical',
'Name': 'hp1',
'Range': {
'CategoricalParameterRangeSpecification': {
'Values': ['TF', 'MXNet']
}
},
'IsTunable': True,
'IsRequired': True,
}, {
'Name': 'hp2',
'Description': 'A categorical hyperparameter',
'Type': 'Categorical',
'IsTunable': False,
'IsRequired': True
}, {
'Name': 'free_text_hp1',
'Description': 'You can write anything here',
'Type': 'FreeText',
'IsTunable': False,
'IsRequired': True
}]
some_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
some_algo['TrainingSpecification'][
'SupportedHyperParameters'] = hyperparameters
sagemaker_session.sagemaker_client.describe_algorithm = Mock(
return_value=some_algo)
estimator = AlgorithmEstimator(
algorithm_arn=
'arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees',
role='SageMakerRole',
train_instance_type='ml.m4.2xlarge',
train_instance_count=1,
sagemaker_session=sagemaker_session,
)
# All 3 Hyperparameters are provided
estimator.set_hyperparameters(hp1='TF', hp2='TF2', free_text_hp1='Hello!')
def test_algorithm_supported_with_spot_instances(session):
session.sagemaker_client.describe_algorithm = Mock(return_value=DESCRIBE_ALGORITHM_RESPONSE)
assert AlgorithmEstimator(
algorithm_arn="arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
instance_type="ml.m4.xlarge",
instance_count=1,
use_spot_instances=True,
max_wait=500,
sagemaker_session=session,
)
def test_algorithm_hyperparameter_integer_range_invalid_range(session):
hyperparameters = [
{
"Description": "Grow a tree with max_leaf_nodes in best-first fashion.",
"Type": "Integer",
"Name": "max_leaf_nodes",
"Range": {
"IntegerParameterRangeSpecification": {"MinValue": "1", "MaxValue": "100000"}
},
"IsTunable": True,
"IsRequired": False,
"DefaultValue": "100",
}
]
some_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
some_algo["TrainingSpecification"]["SupportedHyperParameters"] = hyperparameters
session.sagemaker_client.describe_algorithm = Mock(return_value=some_algo)
estimator = AlgorithmEstimator(
algorithm_arn="arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
instance_type="ml.m4.2xlarge",
instance_count=1,
sagemaker_session=session,
)
with pytest.raises(ValueError):
estimator.set_hyperparameters(max_leaf_nodes=0)
with pytest.raises(ValueError):
estimator.set_hyperparameters(max_leaf_nodes=100001)
def test_algorithm_no_required_hyperparameters(sagemaker_session):
some_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
del some_algo['TrainingSpecification']['SupportedHyperParameters']
sagemaker_session.sagemaker_client.describe_algorithm = Mock(return_value=some_algo)
# Calling AlgorithmEstimator() with unset required hyperparameters
# should fail if they are required.
# Pass training and hyperparameters channels. This should work
assert AlgorithmEstimator(
algorithm_arn='arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees',
role='SageMakerRole',
train_instance_type='ml.m4.2xlarge',
train_instance_count=1,
sagemaker_session=sagemaker_session,
)
def test_algorithm_encrypt_inter_container_traffic(session):
response = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
response['encrypt_inter_container_traffic'] = True
session.sagemaker_client.describe_algorithm = Mock(return_value=response)
estimator = AlgorithmEstimator(
algorithm_arn=
'arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees',
role='SageMakerRole',
train_instance_type='ml.m4.xlarge',
train_instance_count=1,
sagemaker_session=session,
encrypt_inter_container_traffic=True)
encrypt_inter_container_traffic = estimator.encrypt_inter_container_traffic
assert encrypt_inter_container_traffic is True
def test_algorithm_required_hyperparameters_not_provided(session):
hyperparameters = [
{
"Description": "A continuous hyperparameter",
"Type": "Categorical",
"Name": "hp1",
"Range": {
"CategoricalParameterRangeSpecification": {
"Values": ["TF", "MXNet"]
}
},
"IsTunable": True,
"IsRequired": True,
},
{
"Name": "hp2",
"Description": "A continuous hyperparameter",
"Type": "Categorical",
"IsTunable": False,
"IsRequired": True,
},
]
some_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
some_algo["TrainingSpecification"][
"SupportedHyperParameters"] = hyperparameters
session.sagemaker_client.describe_algorithm = Mock(return_value=some_algo)
estimator = AlgorithmEstimator(
algorithm_arn=
"arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
train_instance_type="ml.m4.2xlarge",
train_instance_count=1,
sagemaker_session=session,
)
# hp1 is required and was not provided
with pytest.raises(ValueError):
estimator.set_hyperparameters(hp2="TF2")
# Calling fit with unset required hyperparameters should fail
# this covers the use case of not calling set_hyperparameters() explicitly
with pytest.raises(ValueError):
estimator.fit({"training": "s3://some/place"})
def test_algorithm_trainining_channels_with_invalid_channels(
sagemaker_session):
training_channels = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
training_channels['TrainingSpecification']['TrainingChannels'] = [
{
'Name': 'training',
'Description': 'Input channel that provides training data',
'IsRequired': True,
'SupportedContentTypes': ['text/csv'],
'SupportedCompressionTypes': ['None'],
'SupportedInputModes': ['File'],
},
{
'Name': 'validation',
'Description': 'Input channel that provides validation data',
'IsRequired': False,
'SupportedContentTypes': ['text/csv'],
'SupportedCompressionTypes': ['None'],
'SupportedInputModes': ['File'],
},
]
sagemaker_session.sagemaker_client.describe_algorithm = Mock(
return_value=training_channels)
estimator = AlgorithmEstimator(
algorithm_arn=
'arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees',
role='SageMakerRole',
train_instance_type='ml.m4.xlarge',
train_instance_count=1,
sagemaker_session=sagemaker_session,
)
# Passing only validation should fail as training is required.
with pytest.raises(ValueError):
estimator.fit({'validation': 's3://some/thing'})
# Passing an unknown channel should fail???
with pytest.raises(ValueError):
estimator.fit({
'training': 's3://some/data',
'training2': 's3://some/other/data'
})
def test_algorithm_required_hyperparameters_not_provided(sagemaker_session):
hyperparameters = [{
'Description': 'A continuous hyperparameter',
'Type': 'Categorical',
'Name': 'hp1',
'Range': {
'CategoricalParameterRangeSpecification': {
'Values': ['TF', 'MXNet']
}
},
'IsTunable': True,
'IsRequired': True,
}, {
'Name': 'hp2',
'Description': 'A continuous hyperparameter',
'Type': 'Categorical',
'IsTunable': False,
'IsRequired': True
}]
some_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
some_algo['TrainingSpecification'][
'SupportedHyperParameters'] = hyperparameters
sagemaker_session.sagemaker_client.describe_algorithm = Mock(
return_value=some_algo)
estimator = AlgorithmEstimator(
algorithm_arn=
'arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees',
role='SageMakerRole',
train_instance_type='ml.m4.2xlarge',
train_instance_count=1,
sagemaker_session=sagemaker_session,
)
# hp1 is required and was not provided
with pytest.raises(ValueError):
estimator.set_hyperparameters(hp2='TF2')
# Calling fit with unset required hyperparameters should fail
# this covers the use case of not calling set_hyperparameters() explicitly
with pytest.raises(ValueError):
estimator.fit({'training': 's3://some/place'})
def test_algorithm_trainining_channels_with_invalid_channels(session):
training_channels = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
training_channels["TrainingSpecification"]["TrainingChannels"] = [
{
"Name": "training",
"Description": "Input channel that provides training data",
"IsRequired": True,
"SupportedContentTypes": ["text/csv"],
"SupportedCompressionTypes": ["None"],
"SupportedInputModes": ["File"],
},
{
"Name": "validation",
"Description": "Input channel that provides validation data",
"IsRequired": False,
"SupportedContentTypes": ["text/csv"],
"SupportedCompressionTypes": ["None"],
"SupportedInputModes": ["File"],
},
]
session.sagemaker_client.describe_algorithm = Mock(
return_value=training_channels)
estimator = AlgorithmEstimator(
algorithm_arn=
"arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
train_instance_type="ml.m4.xlarge",
train_instance_count=1,
sagemaker_session=session,
)
# Passing only validation should fail as training is required.
with pytest.raises(ValueError):
estimator.fit({"validation": "s3://some/thing"})
# Passing an unknown channel should fail???
with pytest.raises(ValueError):
estimator.fit({
"training": "s3://some/data",
"training2": "s3://some/other/data"
})
def test_algorithm_trainining_channels_with_expected_channels(
sagemaker_session):
training_channels = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
training_channels['TrainingSpecification']['TrainingChannels'] = [
{
'Name': 'training',
'Description': 'Input channel that provides training data',
'IsRequired': True,
'SupportedContentTypes': ['text/csv'],
'SupportedCompressionTypes': ['None'],
'SupportedInputModes': ['File'],
},
{
'Name': 'validation',
'Description': 'Input channel that provides validation data',
'IsRequired': False,
'SupportedContentTypes': ['text/csv'],
'SupportedCompressionTypes': ['None'],
'SupportedInputModes': ['File'],
},
]
sagemaker_session.sagemaker_client.describe_algorithm = Mock(
return_value=training_channels)
estimator = AlgorithmEstimator(
algorithm_arn=
'arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees',
role='SageMakerRole',
train_instance_type='ml.m4.xlarge',
train_instance_count=1,
sagemaker_session=sagemaker_session,
)
# Pass training and validation channels. This should work
estimator.fit({
'training': 's3://some/place',
'validation': 's3://some/other'
})
# Passing only the training channel. Validation is optional so this should also work.
estimator.fit({'training': 's3://some/place'})
def test_algorithm_trainining_channels_with_expected_channels(session):
training_channels = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
training_channels["TrainingSpecification"]["TrainingChannels"] = [
{
"Name": "training",
"Description": "Input channel that provides training data",
"IsRequired": True,
"SupportedContentTypes": ["text/csv"],
"SupportedCompressionTypes": ["None"],
"SupportedInputModes": ["File"],
},
{
"Name": "validation",
"Description": "Input channel that provides validation data",
"IsRequired": False,
"SupportedContentTypes": ["text/csv"],
"SupportedCompressionTypes": ["None"],
"SupportedInputModes": ["File"],
},
]
session.sagemaker_client.describe_algorithm = Mock(
return_value=training_channels)
estimator = AlgorithmEstimator(
algorithm_arn=
"arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
train_instance_type="ml.m4.xlarge",
train_instance_count=1,
sagemaker_session=session,
)
# Pass training and validation channels. This should work
estimator.fit({
"training": "s3://some/place",
"validation": "s3://some/other"
})
# Passing only the training channel. Validation is optional so this should also work.
estimator.fit({"training": "s3://some/place"})
def test_algorithm_hyperparameter_integer_range_invalid_range(session):
hyperparameters = [{
'Description':
'Grow a tree with max_leaf_nodes in best-first fashion.',
'Type': 'Integer',
'Name': 'max_leaf_nodes',
'Range': {
'IntegerParameterRangeSpecification': {
'MinValue': '1',
'MaxValue': '100000'
}
},
'IsTunable': True,
'IsRequired': False,
'DefaultValue': '100',
}]
some_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
some_algo['TrainingSpecification'][
'SupportedHyperParameters'] = hyperparameters
session.sagemaker_client.describe_algorithm = Mock(return_value=some_algo)
estimator = AlgorithmEstimator(
algorithm_arn=
'arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees',
role='SageMakerRole',
train_instance_type='ml.m4.2xlarge',
train_instance_count=1,
sagemaker_session=session,
)
with pytest.raises(ValueError):
estimator.set_hyperparameters(max_leaf_nodes=0)
with pytest.raises(ValueError):
estimator.set_hyperparameters(max_leaf_nodes=100001)
def test_algorithm_supported_input_mode_with_bad_input_types(sagemaker_session):
# verify that the Estimator verifies raises exceptions when
# attempting to train with an incorrect input type
file_mode_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
file_mode_algo['TrainingSpecification']['TrainingChannels'] = [
{
'Name': 'training',
'Description': 'Input channel that provides training data',
'IsRequired': True,
'SupportedContentTypes': ['text/csv'],
'SupportedCompressionTypes': ['None'],
'SupportedInputModes': ['File'],
},
{
'Name': 'validation',
'Description': 'Input channel that provides validation data',
'IsRequired': False,
'SupportedContentTypes': ['text/csv'],
'SupportedCompressionTypes': ['None'],
'SupportedInputModes': ['File', 'Pipe'],
},
]
sagemaker_session.sagemaker_client.describe_algorithm = Mock(return_value=file_mode_algo)
# Creating a Pipe mode Estimator with a File mode algorithm should fail.
with pytest.raises(ValueError):
AlgorithmEstimator(
algorithm_arn='arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees',
role='SageMakerRole',
train_instance_type='ml.m4.xlarge',
train_instance_count=1,
input_mode='Pipe',
sagemaker_session=sagemaker_session,
)
pipe_mode_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
pipe_mode_algo['TrainingSpecification']['TrainingChannels'] = [
{
'Name': 'training',
'Description': 'Input channel that provides training data',
'IsRequired': True,
'SupportedContentTypes': ['text/csv'],
'SupportedCompressionTypes': ['None'],
'SupportedInputModes': ['Pipe'],
},
{
'Name': 'validation',
'Description': 'Input channel that provides validation data',
'IsRequired': False,
'SupportedContentTypes': ['text/csv'],
'SupportedCompressionTypes': ['None'],
'SupportedInputModes': ['File', 'Pipe'],
},
]
sagemaker_session.sagemaker_client.describe_algorithm = Mock(return_value=pipe_mode_algo)
# Creating a File mode Estimator with a Pipe mode algorithm should fail.
with pytest.raises(ValueError):
AlgorithmEstimator(
algorithm_arn='arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees',
role='SageMakerRole',
train_instance_type='ml.m4.xlarge',
train_instance_count=1,
sagemaker_session=sagemaker_session,
)
def test_algorithm_attach_from_hyperparameter_tuning():
session = Mock()
job_name = "training-job-that-is-part-of-a-tuning-job"
algo_arn = "arn:aws:sagemaker:us-east-2:000000000000:algorithm/scikit-decision-trees"
role_arn = "arn:aws:iam::123412341234:role/SageMakerRole"
instance_count = 1
instance_type = "ml.m4.xlarge"
train_volume_size = 30
input_mode = "File"
session.sagemaker_client.list_tags.return_value = {"Tags": []}
session.sagemaker_client.describe_algorithm.return_value = DESCRIBE_ALGORITHM_RESPONSE
session.sagemaker_client.describe_training_job.return_value = {
"TrainingJobName":
job_name,
"TrainingJobArn":
"arn:aws:sagemaker:us-east-2:123412341234:training-job/%s" % job_name,
"TuningJobArn":
"arn:aws:sagemaker:us-east-2:123412341234:hyper-parameter-tuning-job/%s"
% job_name,
"ModelArtifacts": {
"S3ModelArtifacts":
"s3://sagemaker-us-east-2-123412341234/output/model.tar.gz"
},
"TrainingJobOutput": {
"S3TrainingJobOutput":
"s3://sagemaker-us-east-2-123412341234/output/output.tar.gz"
},
"TrainingJobStatus":
"Succeeded",
"HyperParameters": {
"_tuning_objective_metric": "validation:accuracy",
"max_leaf_nodes": 1,
"free_text_hp1": "foo",
},
"AlgorithmSpecification": {
"AlgorithmName": algo_arn,
"TrainingInputMode": input_mode
},
"MetricDefinitions": [{
"Name": "validation:accuracy",
"Regex": "validation-accuracy: (\\S+)"
}],
"RoleArn":
role_arn,
"InputDataConfig": [{
"ChannelName": "training",
"DataSource": {
"S3DataSource": {
"S3DataType": "S3Prefix",
"S3Uri":
"s3://sagemaker-us-east-2-123412341234/input/training.csv",
"S3DataDistributionType": "FullyReplicated",
}
},
"CompressionType": "None",
"RecordWrapperType": "None",
}],
"OutputDataConfig": {
"KmsKeyId": "",
"S3OutputPath": "s3://sagemaker-us-east-2-123412341234/output",
"RemoveJobNameFromS3OutputPath": False,
},
"ResourceConfig": {
"InstanceType": instance_type,
"InstanceCount": instance_count,
"VolumeSizeInGB": train_volume_size,
},
"StoppingCondition": {
"MaxRuntimeInSeconds": 86400
},
}
estimator = AlgorithmEstimator.attach(job_name, sagemaker_session=session)
assert estimator.hyperparameters() == {
"max_leaf_nodes": 1,
"free_text_hp1": "foo"
}
assert estimator.algorithm_arn == algo_arn
assert estimator.role == role_arn
assert estimator.train_instance_count == instance_count
assert estimator.train_instance_type == instance_type
assert estimator.train_volume_size == train_volume_size
assert estimator.input_mode == input_mode
assert estimator.sagemaker_session == session
def test_algorithm_supported_input_mode_with_bad_input_types(session):
# verify that the Estimator verifies raises exceptions when
# attempting to train with an incorrect input type
file_mode_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
file_mode_algo["TrainingSpecification"]["TrainingChannels"] = [
{
"Name": "training",
"Description": "Input channel that provides training data",
"IsRequired": True,
"SupportedContentTypes": ["text/csv"],
"SupportedCompressionTypes": ["None"],
"SupportedInputModes": ["File"],
},
{
"Name": "validation",
"Description": "Input channel that provides validation data",
"IsRequired": False,
"SupportedContentTypes": ["text/csv"],
"SupportedCompressionTypes": ["None"],
"SupportedInputModes": ["File", "Pipe"],
},
]
session.sagemaker_client.describe_algorithm = Mock(
return_value=file_mode_algo)
# Creating a Pipe mode Estimator with a File mode algorithm should fail.
with pytest.raises(ValueError):
AlgorithmEstimator(
algorithm_arn=
"arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
train_instance_type="ml.m4.xlarge",
train_instance_count=1,
input_mode="Pipe",
sagemaker_session=session,
)
pipe_mode_algo = copy.deepcopy(DESCRIBE_ALGORITHM_RESPONSE)
pipe_mode_algo["TrainingSpecification"]["TrainingChannels"] = [
{
"Name": "training",
"Description": "Input channel that provides training data",
"IsRequired": True,
"SupportedContentTypes": ["text/csv"],
"SupportedCompressionTypes": ["None"],
"SupportedInputModes": ["Pipe"],
},
{
"Name": "validation",
"Description": "Input channel that provides validation data",
"IsRequired": False,
"SupportedContentTypes": ["text/csv"],
"SupportedCompressionTypes": ["None"],
"SupportedInputModes": ["File", "Pipe"],
},
]
session.sagemaker_client.describe_algorithm = Mock(
return_value=pipe_mode_algo)
# Creating a File mode Estimator with a Pipe mode algorithm should fail.
with pytest.raises(ValueError):
AlgorithmEstimator(
algorithm_arn=
"arn:aws:sagemaker:us-east-2:1234:algorithm/scikit-decision-trees",
role="SageMakerRole",
train_instance_type="ml.m4.xlarge",
train_instance_count=1,
sagemaker_session=session,
)
