Sagemaker HyperparameterTuner and fixed hyper parameters (StaticHyperParameters)

Question

I used to use this type of hyper parameter (optimisation) specification:

 "OutputDataConfig": {"S3OutputPath": output_path},
    "ResourceConfig": {"InstanceCount": 1, "InstanceType": "ml.m4.xlarge", "VolumeSizeInGB": 3},
    "RoleArn": role_arn,
    "StaticHyperParameters": {
        "objective": "reg:squarederror"
    },
    "StoppingCondition": {"MaxRuntimeInSeconds": 10000}

TBH I do not even know if this is an old way of doing things or a different SDK – very confusing Sagemaker sometimes. Anyway, I want to use this SDK/API instead – more precisely the HyperparameterTuner. How would I specify StaticHyperParameters (e.g. "objective":"quantile")? Simply by not giving this hyperparameter a range and hard coding it? Thanks!

Asked By: cs0815

||

Source

Answer 1

The hyperparameterTuner takes an Estimator object as one of the parameters. You can keep static hyperparameters as part of the estimator something like below

estimator = PyTorch(
    entry_point="mnist.py",
    role=role,
    py_version="py3",
    framework_version="1.8.0",
    instance_count=1,
    instance_type="ml.c5.2xlarge",
    hyperparameters={"epochs": 1, "backend": "gloo"},
)

Once you have the estimator initialized you can pass this to Tuner along with Parameters that has to be tuned as shown below

hyperparameter_ranges = {
    "lr": ContinuousParameter(0.001, 0.1),
    "batch-size": CategoricalParameter([32, 64, 128, 256, 512]),
}
tuner = HyperparameterTuner(
    estimator,
    objective_metric_name,
    hyperparameter_ranges,
    metric_definitions,
    max_jobs=9,
    max_parallel_jobs=3,
    objective_type=objective_type,
)

Please refer this example for a complete solution

https://github.com/aws/amazon-sagemaker-examples/blob/main/hyperparameter_tuning/pytorch_mnist/hpo_pytorch_mnist.ipynb

Answered By: Arun Lokanatha

Sagemaker HyperparameterTuner and fixed hyper parameters (StaticHyperParameters)

Question:

Answers: