Amazon SageMaker Data Wrangler reduces the time it takes to aggregate and prepare data for ML from weeks to minutes. Using SageMaker Data Wrangler, you can simplify the process of data preparation and feature engineering and complete each step of the data preparation workflow, including data selection, cleansing, exploration, and visualization from a single visual interface.
In this tutorial, you will use Amazon SageMaker Data Wrangler to prepare data to train a rental prediction model. You will use a version of the Brazil house rental dataset found in the Kaggle Data Repository. The data consists of thousands of records, each containing thirteen different features including area, rooms, parking, and other attributes. In addition, each record includes the target feature called rent amount. You will upload the data into Amazon Simple Storage Service (Amazon S3), create a new SageMaker Data Wrangler flow, transform the data, check the data for bias, and lastly save the output to Amazon S3 to be used later for ML training.
What you will accomplish
In this guide, you will:
Visualize and analyze data to understand key relationships
Apply transformations to clean up the data and generate new features
Automatically generate notebooks for repeatable data preparation workflows
Step 1: Set up your Amazon SageMaker Studio domain
With Amazon SageMaker, you can deploy a model visually using the console or programmatically using either SageMaker Studio or SageMaker notebooks. In this tutorial, you deploy the model programmatically using a SageMaker Studio notebook, which requires a SageMaker Studio domain.
An AWS account can have only one SageMaker Studio domain per Region. If you already have a SageMaker Studio domain in the US East (N. Virginia) Region, follow the SageMaker Studio setup guide to attach the required AWS IAM policies to your SageMaker Studio account, then skip Step 1, and proceed directly to Step 2.
If you don't have an existing SageMaker Studio domain, continue with Step 1 to run an AWS CloudFormation template that creates a SageMaker Studio domain and adds the permissions required for the rest of this tutorial.
Choose the AWS CloudFormation stack link. This link opens the AWS CloudFormation console and creates your SageMaker Studio domain and a user named studio-user. It also adds the required permissions to your SageMaker Studio account. In the CloudFormation console, confirm that US East (N. Virginia) is the Region displayed in the upper right corner. Stack name should be CFN-SM-IM-Lambda-catalog, and should not be changed. This stack takes about 10 minutes to create all the resources.
This stack assumes that you already have a public VPC set up in your account. If you do not have a public VPC, see VPC with a single public subnet to learn how to create a public VPC.
Select I acknowledge that AWS CloudFormation might create IAM resources, and then choose Create stack.
On the CloudFormation pane, choose Stacks. It takes about 10 minutes for the stack to be created. When the stack is created, the status of the stack changes from CREATE_IN_PROGRESS to CREATE_COMPLETE.
Step 2: Create a new SageMaker Data Wrangler flow
SageMaker accepts data from a wide variety of sources, including Amazon S3, Amazon Athena, Amazon Redshift, Snowflake, Databricks, and SaaS data sources. In this step, you will create a new SageMaker Data Wrangler flow using the Kaggle Brazil house rental dataset stored in Amazon S3. This dataset contains demographic and financial information about homes along with a target column indicating the rental amount of the property.
Enter SageMaker Studio into the console search bar, and then choose SageMaker Studio.
Choose US East (N. Virginia) from the Region dropdown list on the upper right corner of the SageMaker console. Browse to the Getting Started section in the left-hand navigation and then choose Studio. Then select the studio-user profile and then choose the Open Studio button.
Open the SageMaker Studio interface. On the navigation bar, choose Data Wrangler on the left-hand side, and then choose the Import Data button.
Note that you can change the Flow’s compute instance type using the upper right button showing the current Compute instance. You may decide to change the compute instance type based on your scenario’s dataset size and can scale it up or down when your requirements change. For the purposes of this tutorial, you can use the default ml.m5.4xlarge.
Data Import tab, under
Import data, choose
In the S3 URI Path field, enter s3://sagemaker-sample-files/datasets/tabular/brazil_houses/kaggle_brazil_houses_rental_data.csv, and then choose Go. Under Object name, select kaggle_brazil_houses_rental_data.csv.
In the S3 import details panel, note that you can change the default delimiter and the sampling method when necessary. For the purposes of this tutorial, you can use the default comma delimiter and First K sampling method. Then choose Import.
Step 3: Explore the data
In this step, you use SageMaker Data Wrangler to assess and explore the quality of the training dataset for building machine learning models. Use the Data Quality and Insights report feature to understand your dataset quality, and then use the Quick Model feature to estimate the expected prediction quality and the predictive power of the features in your dataset.
When exploring your dataset, begin by using the Data Quality and Insights report to help you quickly understand your dataset, identify possible issues, and focus your attention on the most important areas to improve the data. On the Data flow tab, in the data flow diagram, choose the + icon, then choose Add analysis. Then choose Get data insights.
From the Data Insights pane, choose rent amount as the Target column. Then choose Regression as the Problem type. Then choose Create.
You may need to wait a minute while the report is generated. Once completed, review the Data Quality and Insights report sections to improve the dataset further before building the ML model. For this specific dataset, the Data Insights report has highlighted two possible issues: the first is related to duplicate rows in the dataset and the second is related to possible target leakage such that one feature is highly correlated with the output and may indicate a duplicate of the target rent column. The report can also be downloaded to a PDF file and shared with colleagues on your team.
For further data analysis and exploration, you can create additional analytical artifacts including correlation matrices, histograms, scatter plots, and summary statistics as well as custom visualizations. For example, choose the + icon, then choose Add analysis.
Under the Create analysis panel, for Analysis type, select Histogram and name it RentHistogramByRooms. For X axis, select rooms.
For Color by, select Rent amount.
Choose Preview to generate a histogram of the rent amount field, color-coded by the rooms variable.
Choose Save to save this analysis to the data flow.
Next, to gain higher confidence that the underlying data has some predictive power, we are going to create a Quick Model. Under the Create analysis pane, for Analysis type, choose Quick Model and name it RentQuickModel.
Then for Label, select rental amount and then choose Preview.
The Quick Model may take several minutes to complete, then the pane shows a brief overview of the Random Cut Forest model built and trained with default hyperparameters. The model generated also displays some statistics, including the Mean Square Error (MSE) score and feature importance to help you evaluate the quality of the dataset. Choose Save.
Step 4: Add transformations to the data flow
SageMaker Data Wrangler simplifies data processing by providing a visual interface with which you can add a wide variety of pre-built transformations. You can also write your custom transformations when necessary using SageMaker Data Wrangler. In this step, you change the type of a string column, rename columns, and drop unnecessary columns using the visual editor.
To navigate to the data flow diagram, choose Data flow. On the data flow diagram, choose the + icon, then Add transform
ALL STEPS pane, choose
The first step is following the Data Insights Report recommendations regarding high risk items and removing the duplicate rows. So as the first transform step, choose Manage Rows, and then select the Drop duplicates operation. Then choose Preview and Save.
Second, we are going to remove the dataset features highlighted as possible sources of target leakage and not appropriate for a machine learning model predicting the rental amount. From the ADD TRANSFORM list, choose Manage columns. Then choose Drop column and choose property tax and fire insurance. Choose Preview then Save.
Next, change the data type of the
floor column from
long. Machine learning models can benefit from using numerically typed columns and this step will allow us to perform further processing later on.
Then rename several columns to improve the readability of the input data set and later analysis.
From the ADD TRANSFORM list, choose Manage columns. Then choose Rename column. Then choose bathroom as the input column and bathrooms as the output column. Choose Preview then Save.
Repeat this renaming column process for hoa [originally from hoa (R$)], rent [originally from rent amount (R$)], and total [originally from total (R$)].
Step 5: Add categorical encoding and numeric scaling transformations to data flow
In this step, you encode categorical variables and scale numerical variables. Categorical encoding transforms string data type categories into numerical features. It’s a common preprocessing task because the numerical features can be used in a wide variety of machine learning model types.
In the dataset, the rental property’s animal and furniture classification is represented by various strings. In this step, you convert these string values to a binary representation, 0 or 1.
Under the ALL STEPS pane, choose + Add step. From the ADD TRANSFORM list, choose Encode categorical. SageMaker Data Wrangler provides three transformation types: Ordinal encode, One hot encode, and Similarity encode. Under the ENCODE CATEGORICAL pane, for Transform, use the default Ordinal encode. For Input columns, select animal and furniture. Ignore the Invalid handling strategy box for this tutorial. Choose Preview, then Add.
To scale the numerical columns area and floor, apply a scaler transformation to normalize the distribution of the data in these columns: Under the ALL STEPS pane, Choose + Add step. From the ADD TRANSFORM list, choose Process numeric. For Scaler, select the default option Standard scaler. For Input columns, select area and floor. Choose Preview, and then Add.
Finally, we will follow another recommendation from the Data Insight report and replace the 0s in the Home Owner Association (hoa) feature with NaN because they indicate missing data and should not be treated as valid inputs that might skew the model. Under the ALL STEPS pane, choose + Add step. From the ADD TRANSFORM list, choose Search and edit and then Convert regex to missing. Choose hoa as the Input column, specify 0 as the Pattern. Click Preview, and then choose Add.
Step 6: Check for data bias
In this step, check your data for bias using Amazon SageMaker Clarify, which provides you with greater visibility into your training data and models so you can identify and limit bias and better explain predictions.
Choose Data flow in the upper left to return to the data flow diagram. Choose the + icon, Add analysis. In the Create analysis pane, for Analysis type, select Bias Report. For Analysis name, enter RentalDataBiasReport. For Select the column your model predicts (target), select rent. Then select Threshold as the predicted column type since this is a regression problem. Specify 3000 as the predicted threshold which corresponds to the average of the rent column in the dataset. Then select city as the column to analyze for bias because we are interested in whether the dataset is imbalanced and over-represents some cities instead of others. Then for Choose bias metrics, keep the default selections. Then choose Check for bias and then Save.
After several seconds, SageMaker Clarify generates a report, which shows how the target and feature columns score on a number of bias-related metrics including Class Imbalance (CI) and Difference in Positive Proportions in Labels (DPL). In this case, the data is slightly biased with regards to rents in Sao Paolo (-0.11), and increasingly skewed for the cities of Rio de Janeiro (0.72), Belo Horizonte (0.77), and Porto Alegre (0.78). Based on this report, you might consider a bias remediation method, such as using SageMaker Data Wrangler’s built-in SMOTE transformation. For the purpose of this tutorial, skip the remediation step. Choose Save to save the bias report to the data flow.
Step 7: Review, integrate, and export your data flow
From the Data Flow tab, review your end-to-end data flow graph including the data source, analytical artifacts, and data transformations. You can easily navigate, view, modify, and delete data flow steps iteratively.
Data Wrangler further streamlines the automation process of exporting the output of the data flow to a persistent destination and can orchestrate the schedule of the flow’s execution. First, set the storage destination to Amazon S3.
Then specify the output dataset name (kaggle_brazil_houses_rental_data_dw_processed.csv) and the Amazon S3 location as your preferred S3 bucket. Then choose Add destination.
Lastly, create the scheduled job that will export the data flow output to Amazon S3 by choosing the Create job button from the Data Flow diagram pane, and then choosing Configure job.
Then you can decide on the job instance type, instance count, the job’s IAM security role, and the job schedule.
SageMaker Autopilot Integration
You can also integrate your data flow with SageMaker Autopilot which automates key tasks of training and deploying a machine learning model. From the Data Flow tab, choose the + icon and then choose Train model.
Export and Train to export the Data Wrangler flow and associate its output with the Autopilot Experiment input.
S3 location where the Data Wrangler flow saved the processed input dataset and specify the
target column as
rent for the Autopilot model.
Specify the Autopilot Training method. You can choose Ensembling, Hyperparameter Optimization, or Auto. For the purposes of this tutorial, choose Auto.
Deployment, select the machine learning problem type as
Regression with the object metric as
Confirm the Autopilot Experiment deployment settings and then choose Create experiment. This action launches a SageMaker Autopilot job that inspects the input data, generates and evaluates multiple ML models, and then selects the best model for subsequent deployment according to the desired performance metric (such as MSE in this tutorial). The Autopilot job may take several minutes to run and complete. Autopilot provides full visibility into how the models were selected, trained, and tuned through a visual leaderboard and programmatic APIs. Finally, Autopilot explains how models make predictions using feature attribution and explainability statistics using SageMaker Clarify.
SageMaker Pipeline Integration
Data Wrangler can also be integrated with SageMaker Inference Pipelines to process data at the time of inference, thereby streamlining the steps between data processing and model inference. When you export one or more steps from the data flow to an inference endpoint, Data Wrangler creates a Jupyter notebook that you can use to define, instantiate, customize, run, and manage the inference pipeline. To create the inference endpoint, choose the + next to the final transformation step (Convert regex to missing) and choose Export to, and then choose SageMaker Inference Pipeline (via Jupyter Notebook). Then inspect and run that Jupyter notebook.
You can optionally export your Data Wrangler data flow to a Jupyter notebook to run the flow steps as a SageMaker Processing job.
Step 8: Clean up resources
It is a best practice to delete resources that you are no longer using so that you don't incur unintended charges.
To delete the S3 bucket, do the following:
Open the Amazon S3 console. On the navigation bar, choose Buckets, sagemaker-<your-Region>-<your-account-id>, and then select the checkbox next to data_wrangler_flows. Then, choose Delete.
In the Delete objects dialog box, verify that you have selected the proper object to delete and enter permanently delete into the Permanently delete objects confirmation box.
Once this is complete and the bucket is empty, you can delete the sagemaker-<your-Region>-<your-account-id> bucket by following the same procedure again.
The Data Science kernel used for running the notebook image in this tutorial will accumulate charges until you either stop the kernel or perform the following steps to delete the apps. For more information, see Shut Down Resources in the Amazon SageMaker Developer Guide.
To delete the SageMaker Studio apps, do the following: On the SageMaker Studio console, choose studio-user, and then delete all the apps listed under Apps by choosing Delete app. Wait until the Status changes to Deleted.
If you used an existing SageMaker Studio domain in Step 1, skip the rest of Step 8 and proceed directly to the conclusion section.
If you ran the CloudFormation template in Step 1 to create a new SageMaker Studio domain, continue with the following steps to delete the domain, user, and the resources created by the CloudFormation template.
To open the CloudFormation console, enter CloudFormation into the AWS console search bar, and choose CloudFormation from the search results.
In the CloudFormation pane, choose Stacks. From the status dropdown list, select Active. Under Stack name, choose CFN-SM-IM-Lambda-catalog to open the stack details page.
On the CFN-SM-IM-Lambda-catalog stack details page, choose Delete to delete the stack along with the resources it created in Step 1.
Congratulations! You have completed the
Prepare Training Data for Machine Learning with Minimal Code tutorial.
You have successfully used Amazon SageMaker Data Wrangler to prepare data for training a machine learning model. SageMaker Data Wrangler offers 300+ preconfigured data transformations, such as convert column type, one-hot encoding, impute missing data with mean or median, re-scale columns, and date/time embeddings, so you can transform your data into formats that can be effectively used for models without writing a single line of code.