def complex_experiment_obj(sagemaker_boto_client):
description = "{}-{}".format("description", str(uuid.uuid4()))
boto3.set_stream_logger("", logging.INFO)
# create experiment
experiment_obj_name = name()
experiment_obj = experiment.Experiment.create(
experiment_name=experiment_obj_name, description=description, sagemaker_boto_client=sagemaker_boto_client
)
# create trials
trial_objs = []
trial_name1 = name()
trial_name2 = name()
trial_name3 = name()
next_trial1 = trial.Trial.create(
trial_name=trial_name1, experiment_name=experiment_obj_name, sagemaker_boto_client=sagemaker_boto_client,
)
trial_objs.append(next_trial1)
next_trial2 = trial.Trial.create(
trial_name=trial_name2, experiment_name=experiment_obj_name, sagemaker_boto_client=sagemaker_boto_client,
)
trial_objs.append(next_trial2)
next_trial3 = trial.Trial.create(
trial_name=trial_name3, experiment_name=experiment_obj_name, sagemaker_boto_client=sagemaker_boto_client,
)
trial_objs.append(next_trial3)
# create trial components
trial_component_name = name()
trial_component_obj = trial_component.TrialComponent.create(
trial_component_name=trial_component_name, sagemaker_boto_client=sagemaker_boto_client,
)
# associate the trials with trial components
for trial_obj in trial_objs:
sagemaker_boto_client.associate_trial_component(
TrialName=trial_obj.trial_name, TrialComponentName=trial_component_obj.trial_component_name
)
time.sleep(1.0)
yield experiment_obj
experiment_obj.delete_all(action="--force")
# load experiment and verify experiment got deleted
with pytest.raises(sagemaker_boto_client.exceptions.ResourceNotFound):
sagemaker_boto_client.describe_experiment(ExperimentName=experiment_obj_name)
# load trials and verify trials got deleted
with pytest.raises(sagemaker_boto_client.exceptions.ResourceNotFound):
sagemaker_boto_client.describe_trial(TrialName=trial_name1)
with pytest.raises(sagemaker_boto_client.exceptions.ResourceNotFound):
sagemaker_boto_client.describe_trial(TrialName=trial_name2)
with pytest.raises(sagemaker_boto_client.exceptions.ResourceNotFound):
sagemaker_boto_client.describe_trial(TrialName=trial_name3)
# load trial component and verify trial component got deleted
with pytest.raises(sagemaker_boto_client.exceptions.ResourceNotFound):
sagemaker_boto_client.describe_trial_component(TrialComponentName=trial_component_name)
def training_s3_uri(boto3_session, tempdir, bucket):
s3_client = boto3_session.client("s3")
filepath = os.path.join(tempdir, name())
with open(filepath, "w") as w:
w.write("Hello World!")
key = "sagemaker/training-input/{}".format(name())
s3_client.upload_file(filepath, bucket, key)
yield "s3://{}/{}".format(bucket, key)
s3_client.delete_object(Bucket=bucket, Key=key)
def test_save_load(experiment_obj, sagemaker_boto_client):
experiment_obj_two = experiment.Experiment.load(experiment_name=experiment_obj.experiment_name,
sagemaker_boto_client=sagemaker_boto_client)
assert experiment_obj.experiment_name == experiment_obj_two.experiment_name
assert experiment_obj.description == experiment_obj_two.description
experiment_obj.description = name()
experiment_obj.display_name = name()
experiment_obj.save()
experiment_obj_three = experiment.Experiment.load(experiment_name=experiment_obj.experiment_name,
sagemaker_boto_client=sagemaker_boto_client)
assert experiment_obj.description == experiment_obj_three.description
assert experiment_obj.display_name == experiment_obj_three.display_name
def trial_component_obj(sagemaker_boto_client):
trial_component_obj = trial_component.TrialComponent.create(
trial_component_name=name(), sagemaker_boto_client=sagemaker_boto_client, tags=TAGS,
)
yield trial_component_obj
time.sleep(0.5)
trial_component_obj.delete()
def test_create_display_name(sagemaker_boto_client):
display_name = name()
tracker_obj = tracker.Tracker.create(display_name=display_name, sagemaker_boto_client=sagemaker_boto_client)
try:
assert display_name == tracker_obj.trial_component.display_name
finally:
tracker_obj.trial_component.delete()
def training_job_name(sagemaker_boto_client, training_role_arn, docker_image,
training_s3_uri, training_output_s3_uri):
training_job_name = name()
sagemaker_boto_client.create_training_job(
TrainingJobName=training_job_name,
InputDataConfig=[{
'ChannelName': 'train',
'DataSource': {
'S3DataSource': {
'S3Uri': training_s3_uri,
'S3DataType': 'S3Prefix'
}
}
}],
AlgorithmSpecification={
'TrainingImage': docker_image,
'TrainingInputMode': 'File',
'EnableSageMakerMetricsTimeSeries': True
},
RoleArn=training_role_arn,
ResourceConfig={
'InstanceType': 'ml.m5.large',
'InstanceCount': 1,
'VolumeSizeInGB': 10
},
StoppingCondition={'MaxRuntimeInSeconds': 900},
OutputDataConfig={'S3OutputPath': training_output_s3_uri})
return training_job_name
def training_job_name(sagemaker_boto_client, training_role_arn, docker_image,
training_s3_uri, training_output_s3_uri):
training_job_name = name()
sagemaker_boto_client.create_training_job(
TrainingJobName=training_job_name,
InputDataConfig=[{
"ChannelName": "train",
"DataSource": {
"S3DataSource": {
"S3Uri": training_s3_uri,
"S3DataType": "S3Prefix"
}
},
}],
AlgorithmSpecification={
"TrainingImage": docker_image,
"TrainingInputMode": "File",
},
RoleArn=training_role_arn,
ResourceConfig={
"InstanceType": "ml.m5.large",
"InstanceCount": 1,
"VolumeSizeInGB": 10
},
StoppingCondition={"MaxRuntimeInSeconds": 900},
OutputDataConfig={"S3OutputPath": training_output_s3_uri},
)
return training_job_name
def trial_obj(sagemaker_boto_client, experiment_obj):
trial_obj = trial.Trial.create(
trial_name=name(),
experiment_name=experiment_obj.experiment_name,
sagemaker_boto_client=sagemaker_boto_client)
yield trial_obj
time.sleep(0.5)
trial_obj.delete()
def experiment_obj(sagemaker_boto_client):
description = '{}-{}'.format('description', str(uuid.uuid4()))
boto3.set_stream_logger('', logging.INFO)
experiment_obj = experiment.Experiment.create(
experiment_name=name(),
description=description,
sagemaker_boto_client=sagemaker_boto_client)
yield experiment_obj
time.sleep(0.5)
experiment_obj.delete()
def processing_job_name(sagemaker_boto_client, training_role_arn, docker_image):
processing_job_name = name()
sagemaker_boto_client.create_processing_job(
ProcessingJobName=processing_job_name,
ProcessingResources={
"ClusterConfig": {"InstanceCount": 1, "InstanceType": "ml.m5.large", "VolumeSizeInGB": 10}
},
AppSpecification={"ImageUri": docker_image},
RoleArn=training_role_arn,
)
return processing_job_name
def trial_component_with_force_disassociation_obj(trials, sagemaker_boto_client):
trial_component_obj = trial_component.TrialComponent.create(
trial_component_name=name(), sagemaker_boto_client=sagemaker_boto_client
)
for trial in trials:
sagemaker_boto_client.associate_trial_component(
TrialName=trial.trial_name, TrialComponentName=trial_component_obj.trial_component_name
)
yield trial_component_obj
time.sleep(0.5)
trial_component_obj.delete(force_disassociate=True)
def experiment_obj(sagemaker_boto_client):
description = "{}-{}".format("description", str(uuid.uuid4()))
boto3.set_stream_logger("", logging.INFO)
experiment_name = name()
experiment_obj = experiment.Experiment.create(
experiment_name=experiment_name, description=description, sagemaker_boto_client=sagemaker_boto_client, tags=TAGS
)
yield experiment_obj
time.sleep(0.5)
experiment_obj.delete()
with pytest.raises(sagemaker_boto_client.exceptions.ResourceNotFound):
sagemaker_boto_client.describe_experiment(ExperimentName=experiment_name)
def processing_job_name(sagemaker_boto_client, training_role_arn,
docker_image):
processing_job_name = name()
sagemaker_boto_client.create_processing_job(
ProcessingJobName=processing_job_name,
ProcessingResources={
'ClusterConfig': {
'InstanceCount': 1,
'InstanceType': 'ml.m5.large',
'VolumeSizeInGB': 10
}
},
AppSpecification={'ImageUri': docker_image},
RoleArn=training_role_arn)
return processing_job_name
def test_log_artifact(trial_component_obj, bucket, tempdir, sagemaker_boto_client):
prefix = name()
file_contents = 'happy monkey monkey'
file_path = os.path.join(tempdir, 'foo.txt')
artifact_name = 'monkey-monkey'
with open(file_path, 'w') as foo_file:
foo_file.write(file_contents)
with tracker.Tracker.load(trial_component_obj.trial_component_name,
artifact_bucket=bucket, artifact_prefix=prefix,
sagemaker_boto_client=sagemaker_boto_client) as tracker_obj:
tracker_obj.log_artifact(file_path, name=artifact_name)
loaded = trial_component.TrialComponent.load(trial_component_name=trial_component_obj.trial_component_name,
sagemaker_boto_client=sagemaker_boto_client)
assert 'text/plain' == loaded.output_artifacts[artifact_name].media_type
assert prefix in loaded.output_artifacts[artifact_name].value
def test_save(experiment_obj):
description = name()
experiment_obj.description = description
experiment_obj.save()
