gluonts.nursery.sagemaker_sdk.estimator module¶
-
class
gluonts.nursery.sagemaker_sdk.estimator.
GluonTSFramework
(sagemaker_session: sagemaker.session.Session, role: str, image_uri: str, base_job_name: str, instance_type: str = 'ml.c5.xlarge', instance_count: int = 1, dependencies: Optional[List[str]] = None, output_path: str = None, code_location: str = None, framework_version: str = '0.4.1', hyperparameters: Dict = None, entry_point: str = '/home/runner/work/gluon-ts/gluon-ts/src/gluonts/nursery/sagemaker_sdk/entry_point_scripts/train_entry_point.py', **kwargs)[source]¶ Bases:
sagemaker.estimator.Framework
This
Estimator
can be used to easily train and evaluate any GluonTS model on any dataset (own or built-in) in AWS Sagemaker using the provided Docker container. It also allows for the execution of custom scripts on AWS Sagemaker. Training is started by callingGluonTSFramework.train()
on this Estimator. After training is complete, callingdeploy()
creates a hosted SageMaker endpoint and returns anGluonTSPredictor
instance that can be used to perform inference against the hosted model. Alternatively, one can call theGluonTSFramework.run()
method to run a custom script defined by the “entry_point” argument of theGluonTSFramework.run()
method. Technical documentation on preparing GluonTSFramework scripts for SageMaker training and using the GluonTsFramework Estimator is available on the project home-page: https://github.com/awslabs/gluon-ts. See how_to_notebooks for examples of how to use this SDK.- Parameters
sagemaker_session – Session object which manages interactions with Amazon SageMaker APIs and any other AWS services needed.
role – An AWS IAM role (either name or full ARN). The Amazon SageMaker training jobs and APIs that create Amazon SageMaker endpoints use this role to access training data and model artifacts. After the endpoint is created, the inference code might use the IAM role, if it needs to access an AWS resource.
image_uri –
The estimator will use this image for training and hosting. It must be an ECR url. If you use an image with MXNET with GPU support, you will have to use a GPU instance. Example:
'123.dkr.ecr.us-west-2.amazonaws.com/my-custom-image:1.0'
base_job_name – Prefix for training job name when the
GluonTSFramework.train()
orGluonTSFramework.run()
method is called.instance_type –
Type of EC2 instance to use for training. Example:
'ml.c5.xlarge' # CPU, 'ml.p2.xlarge' # GPU
instance_count – Currently not more than one supported. Otherwise the number of Amazon EC2 instances to use for training.
dependencies –
A list of paths to files or directories (absolute or relative) with any additional libraries that will be exported to the container. The library folders will be copied to SageMaker in the same folder where the “train.py” is copied. Include a path to a “requirements.txt” to install further dependencies at runtime. The provided dependencies take precedence over the pre-installed ones. If ‘git_config’ is provided, ‘dependencies’ should be a list of relative locations to directories with any additional libraries needed in the Git repo. Example:
GluonTSFramework( entry_point='train.py', dependencies=['my/libs/common', 'requirements.txt'] )
results in the following inside the container:
opt/ml/code ├---> train.py ├---> common └---> requirements.txt
To use a custom GluonTS version just import your custom GluonTS version and then call:
GluonTSFramework( entry_point='train.py', dependencies=[gluonts.__path__[0]] )
This may brake the
GluonTSFramework.train()
method though. If not specified, them dependencies from the Estimator will be used.output_path – S3 location for saving the transform result. If not specified, results are stored to a default bucket.
code_location – The S3 prefix URI where custom code will be uploaded. The code file uploaded in S3 is ‘code_location/source/sourcedir.tar.gz’. If not specified, the default code location is s3://default_bucket/job-name/. And code file uploaded to S3 is s3://default_bucket/job-name/source/sourcedir.tar.gz
framework_version – GluonTS version. If not specified, this will default to 0.4.1. Currently has no effect.
hyperparameters – # TODO add support for HPO Not the Estimator hyperparameters, those are provided through the Estimator in the
GluonTSFramework.train()
method. If you use theGluonTSFramework.run()
method its up to you what you do with this parameter and you could use it to define the hyperparameters of your models. There is no support for Hyper Parameter Optimization (HPO) so far. In general hyperparameters will be used for training. They are made accessible as a dict[str, str] to the training code on SageMaker. For convenience, this accepts other types for keys and values, butstr()
will be called to convert them before training.entry_point – Should not be overwritten if you intend to use the
GluonTSFramework.train()
method, and only be specified through theGluonTSFramework.run()
method.**kwargs – Additional kwargs passed to the
Framework
constructor.
-
LATEST_VERSION
= '0.4.1'¶
-
create_model
(model_server_workers: Optional[str] = None, role: str = None, vpc_config_override: Optional[Dict[str, List[str]]] = 'VPC_CONFIG_DEFAULT', entry_point: str = None, source_dir: str = None, dependencies: List[str] = None, image_uri: str = None, **kwargs) → gluonts.nursery.sagemaker_sdk.model.GluonTSModel[source]¶ Create a
GluonTSModel
object that can be deployed to anEndpoint
.- Parameters
model_server_workers – The number of worker processes used by the inference server. If None, server will use one worker per vCPU.
role – An AWS IAM role (either name or full ARN). The Amazon SageMaker training jobs and APIs that create Amazon SageMaker endpoints use this role to access training data and model artifacts. After the endpoint is created, the inference code might use the IAM role, if it needs to access an AWS resource. If not specified, the role from the Estimator will be used.
vpc_config_override – Optional override for VpcConfig set on the model. Default: use subnets and security groups from this Estimator. * ‘Subnets’ (list[str]): List of subnet ids. * ‘SecurityGroupIds’ (list[str]): List of security group ids.
entry_point – Should not be overwritten if you intend to use the
GluonTSFramework.train()
method, and only be specified through theGluonTSFramework.run()
method.source_dir –
If you set this, your training script will have to be located within the specified source_dir and you will have to set entry_point to the relative path within your source_dir.
Path (absolute, relative, or an S3 URI) to a directory with all training source code including dependencies. Structure within this directory is preserved when training on Amazon SageMaker. If ‘git_config’ is provided, ‘source_dir’ should be a relative location to a directory in the Git repo. For example with the following GitHub repo directory structure:
|---> README.md └---> src |---> train.py └---> test.py
and you need ‘train.py’ as entry point and ‘test.py’ as training source code as well, you must set entry_point=’train.py’, source_dir=’src’. If not specified, the model source directory from training is used.
dependencies –
A list of paths to files or directories (absolute or relative) with any additional libraries that will be exported to the container. The library folders will be copied to SageMaker in the same folder where the “train.py” is copied. Include a path to a “requirements.txt” to install further dependencies at runtime. The provided dependencies take precedence over the pre-installed ones. If ‘git_config’ is provided, ‘dependencies’ should be a list of relative locations to directories with any additional libraries needed in the Git repo. Example:
GluonTSFramework( entry_point='train.py', dependencies=['my/libs/common', 'requirements.txt'] )
results in the following inside the container:
opt/ml/code ├---> train.py ├---> common └---> requirements.txt
To use a custom GluonTS version just import your custom GluonTS version and then call:
GluonTSFramework( entry_point='train.py', dependencies=[gluonts.__path__[0]] )
This may brake the
GluonTSFramework.train()
method though. If not specified, them dependencies from the Estimator will be used.image_uri –
The estimator will use this image for training and hosting. It must be an ECR url. If you use an image with MXNET with GPU support, you will have to use a GPU instance. Example:
'123.dkr.ecr.us-west-2.amazonaws.com/my-custom-image:1.0' 'custom-image:latest'
If not specified, them image from the Estimator will be used.
**kwargs – Additional kwargs passed to the GluonTSModel constructor.
- Returns
A
GluonTSModel
object. SeeGluonTSModel()
for full details.- Return type
gluonts.sagemaker.GluonTSModel
-
classmethod
run
(entry_point: str, inputs, sagemaker_session: sagemaker.session.Session, role: str, image_uri: str, base_job_name: str, instance_type: str, instance_count: int = 1, dependencies: Optional[List[str]] = [], output_path: str = None, code_location: str = None, framework_version: str = '0.4.1', hyperparameters=None, source_dir: str = None, monitored_metrics: List[str] = ('mean_wQuantileLoss', 'ND', 'RMSE'), wait: bool = True, logs: bool = True, job_name: str = None, **kwargs) → Tuple[sagemaker.estimator.Framework, str][source]¶ Use this function to run a custom script specified in ‘entry_point’ in GluonTSFramework. To access files on s3 specify them in inputs. If you want to access local files you should have specified them in ‘dependencies’ in GluonTSFramework.
- Parameters
entry_point –
Path (absolute or relative) to the local Python source file which should be executed as the entry point to training. This should be compatible with Python 3.6. If ‘git_config’ is provided, ‘entry_point’ should be a relative location to the Python source file in the Git repo. For example with the following GitHub repo directory structure:
|---> README.md └---> src |---> train.py └---> test.py
You can assign entry_point=’src/train.py’.
inputs –
Type is str or dict or sagemaker.TrainingInput, however, cannot be empty! Information about the training data. This can be one of three types;
If (str) the S3 location where training data is saved.
- If (dict[str, str] or dict[str, sagemaker.TrainingInput]) If using multiple
channels for training data, you can specify a dict mapping channel names to strings or
TrainingInput()
objects.
- If (sagemaker.TrainingInput) - channel configuration for S3 data sources that can
provide additional information as well as the path to the training dataset. See
sagemaker.TrainingInput()
for full details.
- If (sagemaker.session.FileSystemInput) - channel configuration for
a file system data source that can provide additional information as well as the path to the training dataset.
Example:
inputs = {'my_dataset': sagemaker.TrainingInput(my_dataset_file, content_type='application/json')} # or inputs = {'my_dataset': my_dataset_dir}
where ‘my_dataset_file’ and ‘my_dataset_dir’ are the relative or absolute paths as strings.
sagemaker_session – Session object which manages interactions with Amazon SageMaker APIs and any other AWS services needed.
role – An AWS IAM role (either name or full ARN). The Amazon SageMaker training jobs and APIs that create Amazon SageMaker endpoints use this role to access training data and model artifacts. After the endpoint is created, the inference code might use the IAM role, if it needs to access an AWS resource.
image_uri –
The estimator will use this image for training and hosting. It must be an ECR url. If you use an image with MXNET with GPU support, you will have to use a GPU instance. Example:
'123.dkr.ecr.us-west-2.amazonaws.com/my-custom-image:1.0'
base_job_name – Prefix for training job name when the
GluonTSFramework.train()
orGluonTSFramework.run()
method is called.instance_type –
Type of EC2 instance to use for training. Example:
'ml.c5.xlarge' # CPU, 'ml.p2.xlarge' # GPU
instance_count – Currently not more than one supported. Otherwise the number of Amazon EC2 instances to use for training.
dependencies –
A list of paths to files or directories (absolute or relative) with any additional libraries that will be exported to the container. The library folders will be copied to SageMaker in the same folder where the “train.py” is copied. Include a path to a “requirements.txt” to install further dependencies at runtime. The provided dependencies take precedence over the pre-installed ones. If ‘git_config’ is provided, ‘dependencies’ should be a list of relative locations to directories with any additional libraries needed in the Git repo. Example:
GluonTSFramework.run(entry_point='train.py', dependencies=['my/libs/common', 'requirements.txt'])
results in the following inside the container:
opt/ml/code ├---> train.py ├---> common └---> requirements.txt
To use a custom GluonTS version just import your custom GluonTS version and then call:
GluonTSFramework.run(entry_point='train.py', dependencies=[gluonts.__path__[0]])
This may brake the
GluonTSFramework.train()
method though. If not specified, them dependencies from the Estimator will be used.output_path – S3 location for saving the transform result. If not specified, results are stored to a default bucket.
code_location – The S3 prefix URI where custom code will be uploaded. The code file uploaded in S3 is ‘code_location/source/sourcedir.tar.gz’. If not specified, the default code location is s3://default_bucket/job-name/. And code file uploaded to S3 is s3://default_bucket/job-name/source/sourcedir.tar.gz
framework_version – GluonTS version. If not specified, this will default to 0.4.1. Currently has no effect.
hyperparameters – Its up to you what you do with this parameter and you could use it to define the hyperparameters of your models. In general hyperparameters will be used for training. They are made accessible as a dict[str, str] to the training code on SageMaker. For convenience, this accepts other types for keys and values, but
str()
will be called to convert them before training.source_dir –
If you set this, your training script will have to be located within the specified source_dir and you will have to set entry_point to the relative path within your source_dir. Path (absolute, relative, or an S3 URI) to a directory with all training source code including dependencies. Structure within this directory is preserved when training on Amazon SageMaker. If ‘git_config’ is provided, ‘source_dir’ should be a relative location to a directory in the Git repo. For example with the following GitHub repo directory structure:
|---> README.md └---> src |---> train.py └---> test.py
and you need ‘train.py’ as entry point and ‘test.py’ as training source code as well, you must set entry_point=’train.py’, source_dir=’src’.
monitored_metrics – Names of the metrics that will be parsed from logs in a one minute interval in order to monitor them in Sagemaker.
wait – Whether the call should wait until the job completes (default: True).
logs – Whether to show the logs produced by the job. Only meaningful when wait is True (default: True).
job_name – Training job name. If not specified, a default job name will be generated, based on the base_job_name and the current timestamp.
- Returns
The GluonTSFramework and the job name of the training job.
- Return type
Tuple[Framework, str]
-
train
(dataset: str, estimator: gluonts.model.estimator.Estimator, num_samples: int = 100, quantiles: List[float] = (0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9), monitored_metrics: List[str] = ('mean_wQuantileLoss', 'ND', 'RMSE'), wait: bool = True, logs: bool = True, job_name: str = None) → Union[gluonts.nursery.sagemaker_sdk.estimator.TrainResult, str][source]¶ Use this function to train and evaluate any GluonTS model on Sagemaker. You need to call this method before you can call ‘deploy’.
- Parameters
dataset –
An s3 path-stype URL to a dataset in GluonTs format, or the name of a provided dataset (see gluonts.dataset.repository.datasets.dataset_recipes.keys()). Required dataset structure:
dataset ├---> train | └--> data.json ├---> test | └--> data.json └--> metadata.json
estimator – The GluonTS estimator that should be trained. If you want to train a custom estimator you must have specified the code location in the dependencies argument of the GLuonTSFramework.
num_samples – The num_samples parameter for the gluonts.evaluation.backtest.make_evaluation_predictions method used for evaluation. (default: (0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9))
quantiles – The quantiles parameter for the gluonts.evaluation.Evaluator used for evaluation. (default: 0.1)
monitored_metrics – Names of the metrics that will be parsed from logs in a one minute interval in order to monitor them in Sagemaker.
wait – Whether the call should wait until the job completes (default: True).
logs – Whether to show the logs produced by the job. Only meaningful when wait is True (default: True).
job_name – Training job name. If not specified, a default job name will be generated, based on the base_job_name and the current timestamp.
- Returns
The job name used during training.
- Return type
job_name
-
class
gluonts.nursery.sagemaker_sdk.estimator.
Locations
(job_name, output_path, code_location)[source]¶ Bases:
tuple
-
property
code_location
¶ Alias for field number 2
-
property
estimator_path
¶
-
property
job_code_location
¶
-
property
job_name
¶ Alias for field number 0
-
property
job_output_path
¶
-
property
model_archive
¶
-
property
output_archive
¶
-
property
output_path
¶ Alias for field number 1
-
property