kubeflow
/
pipelines
mirror of https://github.com/kubeflow/pipelines.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
pipelines/components/aws/sagemaker/hyperparameter_tuning/component.yaml

200 lines
8.2 KiB
YAML
Raw Blame History

name: 'SageMaker - Hyperparameter Tuning'
description: |
Hyperparameter Tuning Jobs in SageMaker
inputs:
- name: region
description: 'The region where the cluster launches.'
- name: job_name
description: 'The name of the tuning job. Must be unique within the same AWS account and AWS region.'
default: ''
type: String
- name: role
description: 'The Amazon Resource Name (ARN) that Amazon SageMaker assumes to perform tasks on your behalf.'
type: String
- name: image
description: 'The registry path of the Docker image that contains the training algorithm.'
default: ''
type: String
- name: algorithm_name
description: 'The name of the algorithm resource to use for the hyperparameter tuning job. Do not specify a value for this if using training image.'
default: ''
type: String
- name: training_input_mode
description: 'The input mode that the algorithm supports. File or Pipe.'
default: 'File'
type: String
- name: metric_definitions
description: 'The dictionary of name-regex pairs specify the metrics that the algorithm emits.'
default: '{}'
type: JsonObject
- name: strategy
description: 'How hyperparameter tuning chooses the combinations of hyperparameter values to use for the training job it launches.'
default: 'Bayesian'
type: String
- name: metric_name
description: 'The name of the metric to use for the objective metric.'
type: String
- name: metric_type
description: 'Whether to minimize or maximize the objective metric.'
type: String
- name: early_stopping_type
description: 'Whether to use early stopping for training jobs launched by the tuning job.'
default: 'Off'
type: String
- name: static_parameters
description: 'The values of hyperparameters that do not change for the tuning job.'
default: '{}'
type: JsonObject
- name: integer_parameters
description: 'The array of IntegerParameterRange objects that specify ranges of integer hyperparameters that you want to search.'
default: '[]'
type: JsonArray
- name: continuous_parameters
description: 'The array of ContinuousParameterRange objects that specify ranges of continuous hyperparameters that you want to search.'
default: '[]'
type: JsonObject
- name: categorical_parameters
description: 'The array of CategoricalParameterRange objects that specify ranges of categorical hyperparameters that you want to search.'
default: '[]'
type: JsonArray
- name: channels
description: 'A list of dicts specifying the input channels. Must have at least one.'
type: JsonArray
- name: output_location
description: 'The Amazon S3 path where you want Amazon SageMaker to store the model artifacts is from the best training job.'
type: String
- name: output_encryption_key
description: 'The AWS KMS key that Amazon SageMaker uses to encrypt the model artifacts.'
default: ''
type: String
- name: instance_type
description: 'The ML compute instance type.'
default: 'ml.m4.xlarge'
type: String
- name: instance_count
description: 'The number of ML compute instances to use in each training job.'
default: '1'
type: Integer
- name: volume_size
description: 'The size of the ML storage volume that you want to provision.'
default: '30'
type: Integer
- name: max_num_jobs
description: 'The maximum number of training jobs that a hyperparameter tuning job can launch.'
type: Integer
- name: max_parallel_jobs
description: 'The maximum number of concurrent training jobs that a hyperparameter tuning job can launch.'
type: Integer
- name: max_run_time
description: 'The maximum run time in seconds per training job.'
default: '86400'
type: Integer
- name: resource_encryption_key
description: 'The AWS KMS key that Amazon SageMaker uses to encrypt data on the storage volume attached to the ML compute instance(s).'
default: ''
type: String
- name: vpc_security_group_ids
description: 'The VPC security group IDs, in the form sg-xxxxxxxx.'
default: ''
type: String
- name: vpc_subnets
description: 'The ID of the subnets in the VPC to which you want to connect your hpo job.'
default: ''
type: String
- name: network_isolation
description: 'Isolates the training container.'
default: 'True'
type: Bool
- name: traffic_encryption
description: 'Encrypts all communications between ML compute instances in distributed training.'
default: 'False'
type: Bool
- name: spot_instance
description: 'Use managed spot training.'
default: 'False'
type: Bool
- name: max_wait_time
description: 'The maximum time in seconds you are willing to wait for a managed spot training job to complete.'
default: '86400'
type: Integer
- name: checkpoint_config
description: 'Dictionary of information about the output location for managed spot training checkpoint data.'
default: '{}'
type: JsonObject
- name: warm_start_type
description: 'Specifies either "IdenticalDataAndAlgorithm" or "TransferLearning"'
default: ''
type: String
- name: parent_hpo_jobs
description: 'List of previously completed or stopped hyperparameter tuning jobs to be used as a starting point.'
default: ''
type: String
- name: endpoint_url
description: 'The endpoint URL for the private link VPC endpoint.'
default: ''
type: String
- name: tags
description: 'Key-value pairs, to categorize AWS resources.'
default: '{}'
type: JsonObject
outputs:
- name: hpo_job_name
description: 'The name of the hyper parameter tuning job'
- name: model_artifact_url
description: 'Model artifacts url'
- name: best_job_name
description: 'Best training job in the hyper parameter tuning job'
- name: best_hyperparameters
description: 'Tuned hyperparameters'
- name: training_image
description: 'The registry path of the Docker image that contains the training algorithm'
implementation:
container:
image: amazon/aws-sagemaker-kfp-components:0.3.1
command: ['python3']
args: [
hyperparameter_tuning.py,
--region, {inputValue: region},
--endpoint_url, {inputValue: endpoint_url},
--job_name, {inputValue: job_name},
--role, {inputValue: role},
--image, {inputValue: image},
--algorithm_name, {inputValue: algorithm_name},
--training_input_mode, {inputValue: training_input_mode},
--metric_definitions, {inputValue: metric_definitions},
--strategy, {inputValue: strategy},
--metric_name, {inputValue: metric_name},
--metric_type, {inputValue: metric_type},
--early_stopping_type, {inputValue: early_stopping_type},
--static_parameters, {inputValue: static_parameters},
--integer_parameters, {inputValue: integer_parameters},
--continuous_parameters, {inputValue: continuous_parameters},
--categorical_parameters, {inputValue: categorical_parameters},
--channels, {inputValue: channels},
--output_location, {inputValue: output_location},
--output_encryption_key, {inputValue: output_encryption_key},
--instance_type, {inputValue: instance_type},
--instance_count, {inputValue: instance_count},
--volume_size, {inputValue: volume_size},
--max_num_jobs, {inputValue: max_num_jobs},
--max_parallel_jobs, {inputValue: max_parallel_jobs},
--resource_encryption_key, {inputValue: resource_encryption_key},
--max_run_time, {inputValue: max_run_time},
--vpc_security_group_ids, {inputValue: vpc_security_group_ids},
--vpc_subnets, {inputValue: vpc_subnets},
--network_isolation, {inputValue: network_isolation},
--traffic_encryption, {inputValue: traffic_encryption},
--spot_instance, {inputValue: spot_instance},
--max_wait_time, {inputValue: max_wait_time},
--checkpoint_config, {inputValue: checkpoint_config},
--warm_start_type, {inputValue: warm_start_type},
--parent_hpo_jobs, {inputValue: parent_hpo_jobs},
--tags, {inputValue: tags}
]
fileOutputs:
hpo_job_name: /tmp/hpo_job_name.txt
model_artifact_url: /tmp/model_artifact_url.txt
best_job_name: /tmp/best_job_name.txt
best_hyperparameters: /tmp/best_hyperparameters.txt
training_image: /tmp/training_image.txt
Reference in New Issue View Git Blame Copy Permalink