How Marketers Can Boost Machine Learning Knowledge with Amazon

Marketers are always challenged to raise their machine learning knowledge but have few options to do so without being overwhelmed by programming code. Amazon recognized this need, seeing the popularity of low-code and no-code platforms.  

Late last year, Amazon introduced a number of features in SageMaker Studio, its online integrated development environment (IDE) that supports the SageMaker service for building, training and deploying machine learning models. Two of those services, Canvas and Studio Lab, offer users a convenient environment for data exploration and machine learning project support without requiring extensive technical knowledge. 

Here is an overview of both services and their role in the growing data model choices for marketing teams:

SageMaker Canvas: Assessing Datasets

The most notable no-code service in the SageMaker suite is SageMaker Canvas, a visual point and click user interface for creating machine learning models.  SageMaker Canvas provides options for accessing datasets and performing explorations, such as data joins and SQL queries prior to data import for more control over the data being ingested. It can train multiple models automatically, returning a visual performance model for prediction analysis.

Canvas accepts CSV files through Amazon S3 or local upload. You can also import from Amazon Redshift, or SnowFlake. After importing the data set, you can explore the data, combining datasets with left, right, inner and outer joins, for example. 

Once you have completed your data exploration, you can select the prediction to be conducted. Canvas will automatically determine the problem type based on the applied data set — binary or multi-class classification or regression. You select the your target variable column that displays the results.  

When you initiate training, Canvas will build up to 250 models and choose the one that performs the best. You can preview the results to see if you feel the “best model” chosen is really a fit for your model objectives. You can examine the estimated accuracy of the model and use it to generate predictions on new datasets.

SageMaker Canvas is meant more for business analysts to drive development while keeping costs low. Its only key competitor is Azure Machine Learning. It has similar no-code features to SageMaker, offering on moveable tiles in the interface to establish a machine learning process.  

Related Article: Machine Learning Trends for Marketers to Watch in 2022

SageMaker Studio Lab: Running ML Projects

When teams need to run a project, they can turn to SageMaker Studio Lab, a hosted service which runs machine learning projects without the need for an AWS account, credit card or extensive cloud configuration skills. This service gives interested technologists access to CPU and GPU resources for data modeling. CPUs and GPUs are computing processing units: CPUs are used, naturally, on laptops for data models, while GPUs are typically cloud services for advanced machine learning modeling.

When you create a Studio Lab account at the Amazon SageMaker site, you are first asked a few questions on your programming needs. As of this writing, Amazon normally places you on an account request waitlist to review the answers. But the wait time is a formality. When I tried to set up an account, I received account access within 24 hours.  

Once in your account, a row of project feature appears onscreen meant to start the runtime for your project. The project is run in a JupyterLab environment.  Jupyter is a popular notebook framework for sharing a set of files of a data model project. 

You then make your processor choice. Two options are available based on the amount of processing capability your data model will consume: A CPU-only back-end or a CPU + GPU back-end. The CPU-only back-end offers 12 hours of compute time, enough for data pre-processing tasks that you use on laptop, and standard ML algorithm training. The GPU back-end choice offers four hours of computation time. You can switch processor choices anytime, from CPU to GPU, or vice versa.

Other interface features may feel overly technical for professionals who are not comfortable with developer staples such as Git and Jupyter. But the platform offers a few noteworthy guides so users can get comfortable with exploring data and model concepts on a trial basis.

Once users feel comfortable with Studio Lab, they can migrate to Studio Lab for a full model setup, have more storage and more production-ready options.

Finding the Right Sandbox for Marketers

Tools like SageMaker Studio Labs and Canvas reflect the current state of the art behind work solutions. As CMSWire reporter David Roe noted in his post, low-code and no-code platforms are quickly becoming a suitable business solution to accelerate application development. Development tools are appearing frequently in the no-code platforms space, but nearly all are aimed at the modeling needs of developer teams.   

Marketers should experiment with these tools where possible. Doing so will help marketing teams be better prepared for developing statistically sound solutions based on first-party data.

Allowing marketing professionals to play in the machine learning space — especially when other marketing operations are increasingly using Python — is how tools like SageMaker Studio Labs and Canvas bring value.