The AWS Step Functions Data Science SDK is an open-source library that allows data scientists to easily create workflows that process and publish machine learning models using Amazon SageMaker and AWS Step Functions. You can create machine learning workflows in Python that orchestrate AWS infrastructure at scale, without having to provision and integrate the AWS services separately.
- Workflow - A sequence of steps designed to perform some work
- Step - A unit of work within a workflow
- ML Pipeline - A type of workflow used in data science to create and train machine learning models
The AWS Step Functions Data Science SDK enables you to do the following.
- Easily construct and run machine learning workflows that use AWS infrastructure directly in Python
- Instantiate common training pipelines
- Create standard machine learning workflows in a Jupyter notebook from templates
- Getting Started With Sample Jupyter Notebooks
- Installing the AWS Step Functions Data Science SDK
- Overview of SDK
- AWS Permissions
- Licensing
- Verifying the Signature
- testing
The best way to quickly review how the AWS Step Functions Data Science SDK works is to review the related example notebooks. These notebooks provide code and descriptions for creating and running workflows in AWS Step Functions Using the AWS Step Functions Data Science SDK.
In Amazon SageMaker, example Jupyter notebooks are available in the example notebooks portion of a notebook instance. To run the example notebooks, do the following.
- Either Create a Notebook Instance or Access an Existing notebook instance.
- Select the SageMaker Examples tab.
- Choose a notebook in the Step Functions Data Science SDK section and select Use.
For more information, see Example Notebooks in the Amazon SageMaker documentation.
To run the AWS Step Functions Data Science SDK example notebooks locally, download the sample notebooks and open them in a working Jupyter instance.
- Install Jupyter: https://jupyter.readthedocs.io/en/latest/install.html
- Download the following files from: https://github.com/awslabs/amazon-sagemaker-examples/tree/master/step-functions-data-science-sdk.
hello_world_workflow.ipynbmachine_learning_workflow_abalone.ipynbtraining_pipeline_pytorch_mnist.ipynb
- Open the files in Jupyter.
The AWS Step Functions Data Science SDK is built to PyPI and can be installed with pip as follows.
pip install stepfunctions
You can install from source by cloning this repository and running a pip install command in the root directory of the repository:
git clone https://github.com/aws/aws-step-functions-data-science-sdk-python.git cd aws-step-functions-data-science-sdk-python pip install .
The AWS Step Functions Data Science SDK supports Unix/Linux and Mac.
The AWS Step Functions Data Science SDK is tested on:
- Python 3.6
The AWS Step Functions Data Science SDK provides a Python API that enables you to create data science and machine learning workflows using AWS Step Functions and SageMaker directly in your Python code and Jupyter notebooks.
Using this SDK you can:
- Create steps that accomplish tasks.
- Chain those steps together into workflows.
- Include retry, succeed, or fail steps.
- Review a graphical representation and definition for your workflow.
- Create a workflow in AWS Step Functions.
- Start and review executions in AWS Step Functions.
For a detailed API reference of the AWS Step Functions Data Science SDK, be sure to view this documentation on Read the Docs.
AWS Step Functions lets you coordinate multiple AWS services into serverless workflows so you can build and update apps quickly. Using Step Functions, you can design and run workflows that combine services such as Amazon SageMaker, AWS Lambda, and Amazon Elastic Container Service (Amazon ECS), into feature-rich applications. Workflows are made up of a series of steps, with the output of one step acting as input to the next.
The AWS Step Functions Data Science SDK provides access to AWS Step Functions so that you can easily create and run machine learning and data science workflows directly in Python, and inside your Jupyter Notebooks. Workflows are created locally in Python, but when they are ready for execution, the workflow is first uploaded to the AWS Step Functions service for execution in the cloud.
When you use the SDK to create, update, or execute workflows you are talking to the Step Functions service in the cloud. Your workflows live in AWS Step Functions and can be re-used.
You can execute a workflow as many times as you want, and you can optionally change the input each time. Each time you execute a workflow, it creates a new execution instance in the cloud. You can inspect these executions with SDK commands, or with the Step Functions management console. You can run more than one execution at a time.
Using this SDK you can create steps, chain them together to create a workflow, create that workflow in AWS Step Functions, and execute the workflow in the AWS cloud.
Once you have created your workflow in AWS Step Functions, you can execute that workflow in Step Functions, in the AWS cloud.
Step Functions creates workflows out of steps called States, and expresses that workflow in the Amazon States Language. When you create a workflow in the AWS Step Functions Data Science SDK, it creates a State Machine representing your workflow and steps in AWS Step Functions.
For more information about Step Functions concepts and use, see the Step Functions documentation.
You create steps using the SDK, and chain them together into sequential workflows. Then, you can create those workflows in AWS Step Functions and execute them in Step Functions directly from your Python code. For example, the following is how you define a pass step.
start_pass_state = Pass(
state_id="MyPassState"
)The following is how you define a wait step.
wait_state = Wait(
state_id="Wait for 3 seconds",
seconds=3
)The following example shows how to define a Lambda step, and then defines a Retry and a Catch.
lambda_state = LambdaStep(
state_id="Convert HelloWorld to Base64",
parameters={
"FunctionName": "MyLambda", #replace with the name of your function
"Payload": {
"input": "HelloWorld"
}
}
)
lambda_state.add_retry(Retry(
error_equals=["States.TaskFailed"],
interval_seconds=15,
max_attempts=2,
backoff_rate=4.0
))
lambda_state.add_catch(Catch(
error_equals=["States.TaskFailed"],
next_step=Fail("LambdaTaskFailed")
))After you define these steps, chain them together into a logical sequence.
workflow_definition=Chain([start_pass_state, wait_state, lambda_state])Once the steps are chained together, you can define the workflow definition.
workflow = Workflow(
name="MyWorkflow_v1234",
definition=workflow_definition,
role=stepfunctions_execution_role
)The following generates a graphical representation of your workflow. Please note that visualization currently only works in Jupyter notebooks. Visualization is not available in JupyterLab.
workflow.render_graph(portrait=False)The following renders the JSON of the Amazon States Language definition of the workflow you created.
print(workflow.definition.to_json(pretty=True))The following creates the workflow in AWS Step Functions.
workflow.create()The following starts an execution of your workflow in AWS Step Functions.
execution = workflow.execute(inputs={
"IsHelloWorldExample": True
})The following generates an AWS CloudFormation Template to deploy your workflow.
get_cloudformation_template()The generated template contains only the StateMachine resource. To reuse the CloudFormation template in a different region, please make sure to update the region specific AWS resources (such as the Lambda ARN and Training Image) in the StateMachine definition.
As a managed service, AWS Step Functions performs operations on your behalf on AWS hardware that is managed by AWS Step Functions. AWS Step Functions can perform only operations that the user permits. You can read more about which permissions are necessary in the AWS Documentation.
The AWS Step Functions Data Science SDK should not require any additional permissions
aside from what is required for using .AWS Step Functions. However, if you are
using an IAM role with a path in it, you should grant permission for
iam:GetRole.
AWS Step Functions Data Science SDK is licensed under the Apache 2.0 License. It is copyright 2019 Amazon.com, Inc. or its affiliates. All Rights Reserved. The license is available at: http://aws.amazon.com/apache2.0/
This section describes the recommended process of verifying the validity of the AWS Data Science Workflows Python SDK's compiled distributions on PyPI.
Whenever you download an application from the internet, we recommend that you authenticate the identity of the software publisher and check that the application is not altered or corrupted since it was published. This protects you from installing a version of the application that contains a virus or other malicious code.
If after running the steps in this topic, you determine that the distribution for the AWS Data Science Workflows Python SDK is altered or corrupted, do NOT install the package. Instead, contact AWS Support (https://aws.amazon.com/contact-us/).
AWS Data Science Workflows Python SDK distributions on PyPI are signed using GnuPG, an open source implementation of the Pretty Good Privacy (OpenPGP) standard for secure digital signatures. GnuPG (also known as GPG) provides authentication and integrity checking through a digital signature. For more information about PGP and GnuPG (GPG), see http://www.gnupg.org.
The first step is to establish trust with the software publisher. Download the public key of the software publisher, check that the owner of the public key is who they claim to be, and then add the public key to your keyring. Your keyring is a collection of known public keys. After you establish the authenticity of the public key, you can use it to verify the signature of the application.
- Installing the GPG Tools
- Authenticating and Importing the Public Key
- Verify the Signature of the Package
If your operating system is Linux or Unix, the GPG tools are likely already installed. To test whether the tools are installed on your system, type gpg at a command prompt. If the GPG tools are installed, you see a GPG command prompt. If the GPG tools are not installed, you see an error stating that the command cannot be found. You can install the GnuPG package from a repository.
To install GPG tools on Debian-based Linux
From a terminal, run the following command: apt-get install gnupg
To install GPG tools on Red Hat–based Linux
From a terminal, run the following command: yum install gnupg
The next step in the process is to authenticate the AWS Data Science Workflows Python SDK public key and add it as a trusted key in your GPG keyring.
To authenticate and import the AWS Data Science Workflows Python SDK public key
1. Copy the key from the following text and paste it into a file called data_science_workflows.key. Make sure to include everything that follows:
-----BEGIN PGP PUBLIC KEY BLOCK-----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=ovUh
-----END PGP PUBLIC KEY BLOCK-----
2. At a command prompt in the directory where you saved data_science_workflows.key, use the following command to import the AWS Data Science Workflows Python SDK public key into your keyring:
gpg --import data_science_workflows.key
The command returns results that are similar to the following:
gpg: key 60EB103AE314A809: public key "Stepfunctions-Python-SDK-Signing <stepfunctions-developer-experience [at] amazon.com>" imported
gpg: Total number processed: 1
gpg: imported: 1
Make a note of the key value; you need it in the next step. In the preceding example, the key value is 60EB103AE314A809.
3. Verify the fingerprint by running the following command, replacing key-value with the value from the preceding step:
gpg --fingerprint <key-value>
This command returns results similar to the following:
pub rsa4096 2019-10-31 [SC] [expires: 2030-10-31] CC16 0577 B7BF 9D3D 6E5D
51C5 60EB 103A E314 A809 uid [ unknown]
Stepfunctions-Python-SDK-Signing
<stepfunctions-developer-experience [at] amazon.com> sub rsa4096 2019-10-31 [E]
[expires: 2030-10-31]
Additionally, the fingerprint string should be identical to CC16 0577 B7BF 9D3D 6E5D 51C5 60EB 103A E314 A809, as shown in the preceding example. Compare the key fingerprint that is returned to the one published on this page. They should match. If they don't match, don't install the AWS Data Science Workflows Python SDK package, and contact AWS Support.
After you install the GPG tools, authenticate and import the AWS Data Science Workflows Python SDK public key, and verify that the public key is trusted, you are ready to verify the signature of the package.
To verify the package signature, do the following.
- Download the detached signature for the package from PyPI
Go to the downloads section for the Data Science Workflows Python SDK https://pypi.org/project/stepfunctions/#files on PyPI, Right-click on the SDK distribution link, and choose "Copy Link Location/Address".
Append the string ".asc" to the end of the link you copied, and paste this new link on your browser.
Your browser will prompt you to download a file, which is the detatched signature associated with the respective distribution. Save the file on your local machine.
2. Verify the signature by running the following command at a command prompt in the directory where you saved signature file and the AWS Data Science Workflows Python SDK installation file. Both files must be present.
gpg --verify <path-to-detached-signature-file>
The output should look something like the following:
gpg: Signature made Thu 31 Oct 12:14:53 2019 PDT
gpg: using RSA key CC160577B7BF9D3D6E5D51C560EB103AE314A809
gpg: Good signature from "Stepfunctions-Python-SDK-Signing <stepfunctions-developer-experience [at] amazon.com>" [unknown]
gpg: WARNING: This key is not certified with a trusted signature!
gpg: There is no indication that the signature belongs to the owner.
Primary key fingerprint: CC16 0577 B7BF 9D3D 6E5D 51C5 60EB 103A E314 A809
If the output contains the phrase Good signature from "AWS Data Science Workflows Python SDK <stepfunctions-developer-experience [at] amazon.com>", it means that the signature has successfully been verified, and you can proceed to run the AWS Data Science Workflows Python SDK package.
If the output includes the phrase BAD signature, check whether you performed the procedure correctly. If you continue to get this response, don't run the installation file that you downloaded previously, and contact AWS Support.
The following are details about the warnings you might see:
WARNING: This key is not certified with a trusted signature! There is no
indication that the signature belongs to the owner. This refers to your
personal level of trust in your belief that you possess an authentic public
key for AWS Data Science Workflows Python SDK. In an ideal world, you would
visit an AWS office and receive the key in person. However, more often you
download it from a website. In this case, the website is an AWS website.
gpg: no ultimately trusted keys found. This means that the specific key is not
"ultimately trusted" by you (or by other people whom you trust).
For more information, see http://www.gnupg.org.