Define a Pipeline

To orchestrate your workflows with Amazon SageMaker Model Building Pipelines, you need to generate a directed acyclic graph (DAG) in the form of a JSON pipeline definition. The following image is a representation of the pipeline DAG that you create in this tutorial:

You can generate your JSON pipeline definition using the SageMaker Python SDK. The following tutorial shows how to generate a pipeline definition for a pipeline that solves a regression problem to determine the age of an abalone based on its physical measurements. For a Jupyter notebook that includes the content in this tutorial that you can run, see Orchestrating Jobs with Amazon SageMaker Model Building Pipelines.

Prerequisites

To run the following tutorial you must do the following:

Set up your notebook instance as outlined in Create a notebook instance. This gives your role permissions to read and write to Amazon S3, and create training, batch transform, and processing jobs in SageMaker.

Grant your notebook permissions to get and pass its own role as shown in Modifying a role permissions policy. Add the following JSON snippet to attach this policy to your role. Replace <your-role-arn> with the ARN used to create your notebook instance.


{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
                "iam:GetRole",
                "iam:PassRole"
            ],
            "Resource": "<your-role-arn>"
        }
    ]
}

Trust the SageMaker service principal by following the steps in Modifying a role trust policy. Add the following statement fragment to the trust relationship of your role:
```
{
      "Sid": "",
      "Effect": "Allow",
      "Principal": {
        "Service": "sagemaker.amazonaws.com"
      },
      "Action": "sts:AssumeRole"
    }
```

Set Up Your Environment

Create a new SageMaker session using the following code block. This returns the role ARN for the session. This role ARN should be the execution role ARN that you set up as a prerequisite.


import boto3
import sagemaker
import sagemaker.session


region = boto3.Session().region_name
sagemaker_session = sagemaker.session.Session()
role = sagemaker.get_execution_role()
default_bucket = sagemaker_session.default_bucket()
model_package_group_name = f"AbaloneModelPackageGroupName"

Create a Pipeline

Run the following steps from your SageMaker notebook instance to create a pipeline including steps for preprocessing, training, evaluation, conditional evaluation, and model registration.

Step 1: Download the Dataset

This notebook uses the UCI Machine Learning Abalone Dataset. The dataset contains the following features:

length – The longest shell measurement of the abalone.
diameter – The diameter of the abalone perpendicular to its length.
height – The height of the abalone with meat in the shell.
whole_weight – The weight of the whole abalone.
shucked_weight – The weight of the meat removed from the abalone.
viscera_weight – The weight of the abalone viscera after bleeding.
shell_weight – The weight of the abalone shell after meat removal and drying.
sex – The sex of the abalone. One of 'M', 'F', or 'I', where 'I' is an infant abalone.
rings – The number of rings in the abalone shell.

The number of rings in the abalone shell is a good approximation for its age using the formula age=rings + 1.5. However, obtaining this number is a time-consuming task. You must cut the shell through the cone, stain the section, and count the number of rings through a microscope. However, the other physical measurements are easier to determine. This notebook uses the dataset to build a predictive model of the variable rings using the other physical measurements.

To download the dataset

Download the dataset into your account's default Amazon S3 bucket.


!mkdir -p data
local_path = "data/abalone-dataset.csv"

s3 = boto3.resource("s3")
s3.Bucket(f"sagemaker-servicecatalog-seedcode-{region}").download_file(
    "dataset/abalone-dataset.csv",
    local_path
)

base_uri = f"s3://{default_bucket}/abalone"
input_data_uri = sagemaker.s3.S3Uploader.upload(
    local_path=local_path, 
    desired_s3_uri=base_uri,
)
print(input_data_uri)

Download a second dataset for batch transformation after your model is created.


local_path = "data/abalone-dataset-batch.csv"

s3 = boto3.resource("s3")
s3.Bucket(f"sagemaker-servicecatalog-seedcode-{region}").download_file(
    "dataset/abalone-dataset-batch",
    local_path
)

base_uri = f"s3://{default_bucket}/abalone"
batch_data_uri = sagemaker.s3.S3Uploader.upload(
    local_path=local_path, 
    desired_s3_uri=base_uri,
)
print(batch_data_uri)

Step 2: Define Pipeline Parameters

This code block defines the following parameters for your pipeline:

processing_instance_count – The instance count of the processing job.
input_data – The Amazon S3 location of the input data.
batch_data – The Amazon S3 location of the input data for batch transformation.
model_approval_status – The approval status to register the trained model with for CI/CD. For more information, see Automate MLOps with SageMaker Projects.


from sagemaker.workflow.parameters import (
    ParameterInteger,
    ParameterString,
)

processing_instance_count = ParameterInteger(
    name="ProcessingInstanceCount",
    default_value=1
)
model_approval_status = ParameterString(
    name="ModelApprovalStatus",
    default_value="PendingManualApproval"
)
input_data = ParameterString(
    name="InputData",
    default_value=input_data_uri,
)
batch_data = ParameterString(
    name="BatchData",
    default_value=batch_data_uri,
)

Step 3: Define a Processing Step for Feature Engineering

This section shows how to create a processing step to prepare the data from the dataset for training.

To create a processing step

Create a directory for the processing script.
```
!mkdir -p abalone
```

Create a file in the /abalone directory named preprocessing.py with the following content. This preprocessing script is passed in to the processing step for execution on the input data. The training step then uses the preprocessed training features and labels to train a model, and the evaluation step uses the trained model and preprocessed test features and labels to evaluate the model. The script uses scikit-learn to do the following:

Fill in missing sex categorical data and encode it so it's suitable for training.
Scale and normalize all numerical fields except for rings and sex.
Split the data into training, test, and validation datasets.


%%writefile abalone/preprocessing.py
import argparse
import os
import requests
import tempfile
import numpy as np
import pandas as pd


from sklearn.compose import ColumnTransformer
from sklearn.impute import SimpleImputer
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler, OneHotEncoder


# Because this is a headerless CSV file, specify the column names here.
feature_columns_names = [
    "sex",
    "length",
    "diameter",
    "height",
    "whole_weight",
    "shucked_weight",
    "viscera_weight",
    "shell_weight",
]
label_column = "rings"

feature_columns_dtype = {
    "sex": str,
    "length": np.float64,
    "diameter": np.float64,
    "height": np.float64,
    "whole_weight": np.float64,
    "shucked_weight": np.float64,
    "viscera_weight": np.float64,
    "shell_weight": np.float64
}
label_column_dtype = {"rings": np.float64}


def merge_two_dicts(x, y):
    z = x.copy()
    z.update(y)
    return z


if __name__ == "__main__":
    base_dir = "/opt/ml/processing"

    df = pd.read_csv(
        f"{base_dir}/input/abalone-dataset.csv",
        header=None, 
        names=feature_columns_names + [label_column],
        dtype=merge_two_dicts(feature_columns_dtype, label_column_dtype)
    )
    numeric_features = list(feature_columns_names)
    numeric_features.remove("sex")
    numeric_transformer = Pipeline(
        steps=[
            ("imputer", SimpleImputer(strategy="median")),
            ("scaler", StandardScaler())
        ]
    )

    categorical_features = ["sex"]
    categorical_transformer = Pipeline(
        steps=[
            ("imputer", SimpleImputer(strategy="constant", fill_value="missing")),
            ("onehot", OneHotEncoder(handle_unknown="ignore"))
        ]
    )

    preprocess = ColumnTransformer(
        transformers=[
            ("num", numeric_transformer, numeric_features),
            ("cat", categorical_transformer, categorical_features)
        ]
    )
    
    y = df.pop("rings")
    X_pre = preprocess.fit_transform(df)
    y_pre = y.to_numpy().reshape(len(y), 1)
    
    X = np.concatenate((y_pre, X_pre), axis=1)
    
    np.random.shuffle(X)
    train, validation, test = np.split(X, [int(.7*len(X)), int(.85*len(X))])

    
    pd.DataFrame(train).to_csv(f"{base_dir}/train/train.csv", header=False, index=False)
    pd.DataFrame(validation).to_csv(f"{base_dir}/validation/validation.csv", header=False, index=False)
    pd.DataFrame(test).to_csv(f"{base_dir}/test/test.csv", header=False, index=False)

Create an instance of an SKLearnProcessor to pass in to the processing step.


from sagemaker.sklearn.processing import SKLearnProcessor


framework_version = "0.23-1"

sklearn_processor = SKLearnProcessor(
    framework_version=framework_version,
    instance_type="ml.m5.xlarge",
    instance_count=processing_instance_count,
    base_job_name="sklearn-abalone-process",
    role=role,
)

Create a processing step. This step takes in the SKLearnProcessor, the input and output channels, and the preprocessing.py script that you created. This is very similar to a processor instance's run method in the SageMaker Python SDK. The input_data parameter passed into ProcessingStep is the input data of the step itself. This input data is used by the processor instance when it runs.

Note the "train, "validation, and "test" named channels specified in the output configuration for the processing job. Step Properties such as these can be used in subsequent steps and resolve to their runtime values at execution.
```
from sagemaker.processing import ProcessingInput, ProcessingOutput
from sagemaker.workflow.steps import ProcessingStep
    

step_process = ProcessingStep(
    name="AbaloneProcess",
    processor=sklearn_processor,
    inputs=[
      ProcessingInput(source=input_data, destination="/opt/ml/processing/input"),  
    ],
    outputs=[
        ProcessingOutput(output_name="train", source="/opt/ml/processing/train"),
        ProcessingOutput(output_name="validation", source="/opt/ml/processing/validation"),
        ProcessingOutput(output_name="test", source="/opt/ml/processing/test")
    ],
    code="abalone/preprocessing.py",
)
```

Step 4: Define a Training step

This section shows how to use the SageMaker XGBoost Algorithm to train a model on the training data output from the processing steps.

To define a training step

Specify the model path where you want to save the models from training.
```
model_path = f"s3://{default_bucket}/AbaloneTrain"
```

Configure an estimator for the XGBoost algorithm and the input dataset. The training instance type is passed into the estimator. A typical training script loads data from the input channels, configures training with hyperparameters, trains a model, and saves a model to model_dir so that it can be hosted later. SageMaker uploads the model to Amazon S3 in the form of a model.tar.gz at the end of the training job.


from sagemaker.estimator import Estimator


image_uri = sagemaker.image_uris.retrieve(
    framework="xgboost",
    region=region,
    version="1.0-1",
    py_version="py3",
    instance_type="ml.m5.xlarge"
)
xgb_train = Estimator(
    image_uri=image_uri,
    instance_type="ml.m5.xlarge",
    instance_count=1,
    output_path=model_path,
    role=role,
)
xgb_train.set_hyperparameters(
    objective="reg:linear",
    num_round=50,
    max_depth=5,
    eta=0.2,
    gamma=4,
    min_child_weight=6,
    subsample=0.7,
    silent=0
)

Create a TrainingStep using the estimator instance and properties of the ProcessingStep. In particular, pass in the S3Uri of the "train" and "validation" output channel to the TrainingStep.


from sagemaker.inputs import TrainingInput
from sagemaker.workflow.steps import TrainingStep


step_train = TrainingStep(
    name="AbaloneTrain",
    estimator=xgb_train,
    inputs={
        "train": TrainingInput(
            s3_data=step_process.properties.ProcessingOutputConfig.Outputs[
                "train"
            ].S3Output.S3Uri,
            content_type="text/csv"
        ),
        "validation": TrainingInput(
            s3_data=step_process.properties.ProcessingOutputConfig.Outputs[
                "validation"
            ].S3Output.S3Uri,
            content_type="text/csv"
        )
    },
)

Step 5: Define a Processing Step for Model Evaluation

This section shows how to create a processing step to evaluate the accuracy of the model. The result of this model evaluation is used in the condition step to determine which execute path to take.

To define a processing step for model evaluation

Create a file in the /abalone directory named evaluation.py. This script is used in a processing step to perform model evaluation. It takes a trained model and the test dataset as input, then produces a JSON file containing classification evaluation metrics.


%%writefile abalone/evaluation.py
import json
import pathlib
import pickle
import tarfile
import joblib
import numpy as np
import pandas as pd
import xgboost


from sklearn.metrics import mean_squared_error


if __name__ == "__main__":
    model_path = f"/opt/ml/processing/model/model.tar.gz"
    with tarfile.open(model_path) as tar:
        tar.extractall(path=".")
    
    model = pickle.load(open("xgboost-model", "rb"))

    test_path = "/opt/ml/processing/test/test.csv"
    df = pd.read_csv(test_path, header=None)
    
    y_test = df.iloc[:, 0].to_numpy()
    df.drop(df.columns[0], axis=1, inplace=True)
    
    X_test = xgboost.DMatrix(df.values)
    
    predictions = model.predict(X_test)

    mse = mean_squared_error(y_test, predictions)
    std = np.std(y_test - predictions)
    report_dict = {
        "regression_metrics": {
            "mse": {
                "value": mse,
                "standard_deviation": std
            },
        },
    }

    output_dir = "/opt/ml/processing/evaluation"
    pathlib.Path(output_dir).mkdir(parents=True, exist_ok=True)
    
    evaluation_path = f"{output_dir}/evaluation.json"
    with open(evaluation_path, "w") as f:
        f.write(json.dumps(report_dict))

Create an instance of a ScriptProcessor that is used to create a ProcessingStep.


from sagemaker.processing import ScriptProcessor


script_eval = ScriptProcessor(
    image_uri=image_uri,
    command=["python3"],
    instance_type="ml.m5.xlarge",
    instance_count=1,
    base_job_name="script-abalone-eval",
    role=role,
)

Create a ProcessingStep using the processor instance, the input and output channels, and the evaluation.py script. In particular, pass in the S3ModelArtifacts property from the step_train training step, as well as the S3Uri of the "test" output channel of the step_process processing step. This is very similar to a processor instance's run method in the SageMaker Python SDK.


from sagemaker.workflow.properties import PropertyFile


evaluation_report = PropertyFile(
    name="EvaluationReport",
    output_name="evaluation",
    path="evaluation.json"
)
step_eval = ProcessingStep(
    name="AbaloneEval",
    processor=script_eval,
    inputs=[
        ProcessingInput(
            source=step_train.properties.ModelArtifacts.S3ModelArtifacts,
            destination="/opt/ml/processing/model"
        ),
        ProcessingInput(
            source=step_process.properties.ProcessingOutputConfig.Outputs[
                "test"
            ].S3Output.S3Uri,
            destination="/opt/ml/processing/test"
        )
    ],
    outputs=[
        ProcessingOutput(output_name="evaluation", source="/opt/ml/processing/evaluation"),
    ],
    code="abalone/evaluation.py",
    property_files=[evaluation_report],
)

Step 6: Define a CreateModelStep for Batch Transformation

Important

We recommend using Model Step to create models as of v2.90.0 of the SageMaker Python SDK. CreateModelStep will continue to work in previous versions of the SageMaker Python SDK, but is no longer actively supported.

This section shows how to create a SageMaker model from the output of the training step. This model is used for batch transformation on a new dataset. This step is passed into the condition step and only executes if the condition step evaluates to true.

To define a CreateModelStep for batch transformation

Create a SageMaker model. Pass in the S3ModelArtifacts property from the step_train training step.


from sagemaker.model import Model


model = Model(
    image_uri=image_uri,
    model_data=step_train.properties.ModelArtifacts.S3ModelArtifacts,
    sagemaker_session=sagemaker_session,
    role=role,
)

Define the model input for your SageMaker model.


from sagemaker.inputs import CreateModelInput


inputs = CreateModelInput(
    instance_type="ml.m5.large",
    accelerator_type="ml.eia1.medium",
)

Create your CreateModelStep using the CreateModelInput and SageMaker model instance you defined.


from sagemaker.workflow.steps import CreateModelStep


step_create_model = CreateModelStep(
    name="AbaloneCreateModel",
    model=model,
    inputs=inputs,
)

Step 7: Define a TransformStep to Perform Batch Transformation

This section shows how to create a TransformStep to perform batch transformation on a dataset after the model is trained. This step is passed into the condition step and only executes if the condition step evaluates to true.

To define a TransformStep to perform batch transformation

Create a transformer instance with the appropriate compute instance type, instance count, and desired output Amazon S3 bucket URI. Pass in the ModelName property from the step_create_model CreateModel step.


from sagemaker.transformer import Transformer


transformer = Transformer(
    model_name=step_create_model.properties.ModelName,
    instance_type="ml.m5.xlarge",
    instance_count=1,
    output_path=f"s3://{default_bucket}/AbaloneTransform"
)

Create a TransformStep using the transformer instance you defined and the batch_data pipeline parameter.


from sagemaker.inputs import TransformInput
from sagemaker.workflow.steps import TransformStep


step_transform = TransformStep(
    name="AbaloneTransform",
    transformer=transformer,
    inputs=TransformInput(data=batch_data)
)

Step 8: Define a RegisterModel Step to Create a Model Package

Important

We recommend using Model Step to register models as of v2.90.0 of the SageMaker Python SDK. RegisterModel will continue to work in previous versions of the SageMaker Python SDK, but is no longer actively supported.

This section shows how to construct an instance of RegisterModel. The result of executing RegisterModel in a pipeline is a model package. A model package is a reusable model artifacts abstraction that packages all ingredients necessary for inference. It consists of an inference specification that defines the inference image to use along with an optional model weights location. A model package group is a collection of model packages. You can use a ModelPackageGroup for SageMaker Pipelines to add a new version and model package to the group for every pipeline execution. For more information about model registry, see Register and Deploy Models with Model Registry.

This step is passed into the condition step and only executes if the condition step evaluates to true.

To define a RegisterModel step to create a model package

Construct a RegisterModel step using the estimator instance you used for the training step . Pass in the S3ModelArtifacts property from the step_train training step and specify a ModelPackageGroup. SageMaker Pipelines creates this ModelPackageGroup for you.


from sagemaker.model_metrics import MetricsSource, ModelMetrics 
from sagemaker.workflow.step_collections import RegisterModel


model_metrics = ModelMetrics(
    model_statistics=MetricsSource(
        s3_uri="{}/evaluation.json".format(
            step_eval.arguments["ProcessingOutputConfig"]["Outputs"][0]["S3Output"]["S3Uri"]
        ),
        content_type="application/json"
    )
)
step_register = RegisterModel(
    name="AbaloneRegisterModel",
    estimator=xgb_train,
    model_data=step_train.properties.ModelArtifacts.S3ModelArtifacts,
    content_types=["text/csv"],
    response_types=["text/csv"],
    inference_instances=["ml.t2.medium", "ml.m5.xlarge"],
    transform_instances=["ml.m5.xlarge"],
    model_package_group_name=model_package_group_name,
    approval_status=model_approval_status,
    model_metrics=model_metrics
)

Step 9: Define a Condition Step to Verify Model Accuracy

A ConditionStep allows SageMaker Pipelines to support conditional execution in your pipeline DAG based on the condition of step properties. In this case, you only want to register a model package if the accuracy of that model, as determined by the model evaluation step, exceeds the required value. If the accuracy exceeds the required value, the pipeline also creates a SageMaker Model and runs batch transformation on a dataset. This section shows how to define the Condition step.

To define a condition step to verify model accuracy

Define a ConditionLessThanOrEqualTo condition using the accuracy value found in the output of the model evaluation processing step, step_eval. Get this output using the property file you indexed in the processing step and the respective JSONPath of the mean squared error value, "mse".


from sagemaker.workflow.conditions import ConditionLessThanOrEqualTo
from sagemaker.workflow.condition_step import ConditionStep
from sagemaker.workflow.functions import JsonGet


cond_lte = ConditionLessThanOrEqualTo(
    left=JsonGet(
        step_name=step_eval.name,
        property_file=evaluation_report,
        json_path="regression_metrics.mse.value"
    ),
    right=6.0
)

Construct a ConditionStep. Pass the ConditionEquals condition in, then set the model package registration and batch transformation steps as the next steps if the condition passes.
```
step_cond = ConditionStep(
    name="AbaloneMSECond",
    conditions=[cond_lte],
    if_steps=[step_register, step_create_model, step_transform],
    else_steps=[], 
)
```

Step 10: Create a pipeline

Now that you’ve created all of the steps, combine them into a pipeline.

To create a pipeline

Define the following for your pipeline: name, parameters, and steps. Names must be unique within an (account, region) pair.

Note

A step can only appear once in either the pipeline's step list or the if/else step lists of the condition step. It cannot appear in both.


from sagemaker.workflow.pipeline import Pipeline


pipeline_name = f"AbalonePipeline"
pipeline = Pipeline(
    name=pipeline_name,
    parameters=[
        processing_instance_count,
        model_approval_status,
        input_data,
        batch_data,
    ],
    steps=[step_process, step_train, step_eval, step_cond],
)

(Optional) Examine the JSON pipeline definition to ensure that it's well-formed.
```
import json

json.loads(pipeline.definition())
```

This pipeline definition is ready to submit to SageMaker. In the next tutorial, you submit this pipeline to SageMaker and start an execution.

Next step: Run a pipeline

Warning Javascript is disabled or is unavailable in your browser.

To use the Amazon Web Services Documentation, Javascript must be enabled. Please refer to your browser's Help pages for instructions.

Document Conventions

Create and Manage Pipelines

Run a pipeline