Skip to content

fix: Allow hyperparameters in Tensorflow estimator to be parameterized #2296

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Apr 22, 2021
Merged

fix: Allow hyperparameters in Tensorflow estimator to be parameterized #2296

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9091973
Allow hyperparameters in Tensorflow estimator to be parameterized for…
Apr 21, 2021
c1ae102
Fix linter errors
Apr 21, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 7 additions & 1 deletion src/sagemaker/estimator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2443,7 +2443,13 @@ def attach(cls, training_job_name, sagemaker_session=None, model_channel_name="m
@staticmethod
def _json_encode_hyperparameters(hyperparameters):
"""Placeholder docstring"""
return {str(k): json.dumps(v) for (k, v) in hyperparameters.items()}
current_hyperparameters = hyperparameters
if current_hyperparameters is not None:
hyperparameters = {
str(k): (v if isinstance(v, (Parameter, Expression, Properties)) else json.dumps(v))
for (k, v) in current_hyperparameters.items()
}
return hyperparameters

@classmethod
def _update_init_params(cls, hp, tf_arguments):
Expand Down
93 changes: 92 additions & 1 deletion tests/unit/sagemaker/workflow/test_steps.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@

import pytest
import sagemaker
import os

from mock import (
Mock,
Expand All @@ -24,6 +25,7 @@

from sagemaker.debugger import ProfilerConfig
from sagemaker.estimator import Estimator
from sagemaker.tensorflow import TensorFlow
from sagemaker.inputs import TrainingInput, TransformInput, CreateModelInput
from sagemaker.model import Model
from sagemaker.processing import (
Expand All @@ -45,6 +47,10 @@
CreateModelStep,
CacheConfig,
)
from tests.unit import DATA_DIR

SCRIPT_FILE = "dummy_script.py"
SCRIPT_PATH = os.path.join(DATA_DIR, SCRIPT_FILE)

REGION = "us-west-2"
BUCKET = "my-bucket"
Expand Down Expand Up @@ -112,7 +118,7 @@ def test_custom_step():
assert step.to_request() == {"Name": "MyStep", "Type": "Training", "Arguments": dict()}


def test_training_step(sagemaker_session):
def test_training_step_base_estimator(sagemaker_session):
instance_type_parameter = ParameterString(name="InstanceType", default_value="c4.4xlarge")
instance_count_parameter = ParameterInteger(name="InstanceCount", default_value=1)
data_source_uri_parameter = ParameterString(
Expand Down Expand Up @@ -177,6 +183,91 @@ def test_training_step(sagemaker_session):
assert step.properties.TrainingJobName.expr == {"Get": "Steps.MyTrainingStep.TrainingJobName"}


def test_training_step_tensorflow(sagemaker_session):
instance_type_parameter = ParameterString(name="InstanceType", default_value="ml.p3.16xlarge")
instance_count_parameter = ParameterInteger(name="InstanceCount", default_value=1)
data_source_uri_parameter = ParameterString(
name="DataSourceS3Uri", default_value=f"s3://{BUCKET}/train_manifest"
)
training_epochs_parameter = ParameterInteger(name="TrainingEpochs", default_value=5)
training_batch_size_parameter = ParameterInteger(name="TrainingBatchSize", default_value=500)
estimator = TensorFlow(
entry_point=os.path.join(DATA_DIR, SCRIPT_FILE),
role=ROLE,
model_dir=False,
image_uri=IMAGE_URI,
source_dir="s3://mybucket/source",
framework_version="2.4.1",
py_version="py37",
instance_count=instance_count_parameter,
instance_type=instance_type_parameter,
sagemaker_session=sagemaker_session,
# subnets=subnets,
hyperparameters={
"batch-size": training_batch_size_parameter,
"epochs": training_epochs_parameter,
},
# security_group_ids=security_group_ids,
debugger_hook_config=False,
# Training using SMDataParallel Distributed Training Framework
distribution={"smdistributed": {"dataparallel": {"enabled": True}}},
)

inputs = TrainingInput(s3_data=data_source_uri_parameter)
cache_config = CacheConfig(enable_caching=True, expire_after="PT1H")
step = TrainingStep(
name="MyTrainingStep", estimator=estimator, inputs=inputs, cache_config=cache_config
)
step_request = step.to_request()
step_request["Arguments"]["HyperParameters"].pop("sagemaker_job_name", None)
step_request["Arguments"]["HyperParameters"].pop("sagemaker_program", None)
step_request["Arguments"].pop("ProfilerRuleConfigurations", None)
assert step_request == {
"Name": "MyTrainingStep",
"Type": "Training",
"Arguments": {
"AlgorithmSpecification": {
"TrainingInputMode": "File",
"TrainingImage": "fakeimage",
"EnableSageMakerMetricsTimeSeries": True,
},
"OutputDataConfig": {"S3OutputPath": "s3://my-bucket/"},
"StoppingCondition": {"MaxRuntimeInSeconds": 86400},
"ResourceConfig": {
"InstanceCount": instance_count_parameter,
"InstanceType": instance_type_parameter,
"VolumeSizeInGB": 30,
},
"RoleArn": "DummyRole",
"InputDataConfig": [
{
"DataSource": {
"S3DataSource": {
"S3DataType": "S3Prefix",
"S3Uri": data_source_uri_parameter,
"S3DataDistributionType": "FullyReplicated",
}
},
"ChannelName": "training",
}
],
"HyperParameters": {
"batch-size": training_batch_size_parameter,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It is better to hardcode the json here to make the assertion stronger:

{"Get": "Parameters.TrainingBatchSize"}

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

At this point it's still dealing with a Parameter object, it hasn't been converted to JSON yet. It gets interpolated when pipeline.definition() is called here, and we have unit tests covering that already

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see.

"epochs": training_epochs_parameter,
"sagemaker_submit_directory": '"s3://mybucket/source"',
"sagemaker_container_log_level": "20",
"sagemaker_region": '"us-west-2"',
"sagemaker_distributed_dataparallel_enabled": "true",
"sagemaker_instance_type": instance_type_parameter,
"sagemaker_distributed_dataparallel_custom_mpi_options": '""',
},
"ProfilerConfig": {"S3OutputPath": "s3://my-bucket/"},
},
"CacheConfig": {"Enabled": True, "ExpireAfter": "PT1H"},
}
assert step.properties.TrainingJobName.expr == {"Get": "Steps.MyTrainingStep.TrainingJobName"}


def test_processing_step(sagemaker_session):
processing_input_data_uri_parameter = ParameterString(
name="ProcessingInputDataUri", default_value=f"s3://{BUCKET}/processing_manifest"
Expand Down