Analytics

News, events, thought leadership and more.
sadovsky
Alteryx Alumni (Retired)

2020.3_AutoML_CommBanner_1000x150px-01.jpg

The Alteryx Intelligence Suite has made it simple for any Alteryx user to unlock the power of advanced machine learning capabilities via the Machine Learning toolkit.  Powered by the best in class open source software for machine learning combined with the ease of use that comes from integration into the Alteryx platform, business users are finding meaningful insights as they better understand the present and start to predict the future from their data.  These users are doing so with full transparency into the process due to our fully detailed, non-blackbox solution. In addition, the recommendations and guidance provided by the software is helping to shepherd a new generation of citizen data scientists.

 

As we’ve talked to our customers, one need has been apparent to us.  While a DIY approach allows exact control over every aspect of the modeling cycle, and the assisted step-by-step format provides an excellent machine automated partner in that journey, users are also interested in fully automated modeling.  The ability to click a button and let a machine try thousands of permutations to find the model that best fits a set of data is attractive to the business analyst, data scientist and anyone in between as it allows them to automate repetitive tasks and reduce the possibility of manual errors.  That’s why we’re excited to announce the new Automatic modeling mode in the 2020.3 Machine Learning building blocks as part of the Intelligence Suite.

 

Our Automated Modeling allows one click model building complete with hyperparameter tuning and other modeling pipeline optimizations while still offering the transparency, control, and workflow efficiencies already offered by the Machine Learning building blocks.  This new approach is seamlessly built into our Assisted Modeling offering and provides models compatible with existing workflows built in the Intelligence Suite. 

 

The growing number of ways that exist to construct predictive models in Alteryx poses a question: which path is the best one to take? Well, that depends. Whether our customers have experience coding or none at all, there are many different ways to solve the same types of predictive questions comfortably in Alteryx, our newest path being Automated Modeling.

 

PredictiveInfographic_600x1200px-01.jpg

 

To highlight some of the amazing capabilities of this new feature, I’m excited to have Josh Herr, one of the engineers I worked with to implement this new way to model, explore this new functionality in detail!

 

Automated Modeling in Detail

We have worked tirelessly to build an automated modeling platform that integrates seamlessly with our current Assisted Modeling system and the Designer application while bringing the full power and utility that automated modeling can provide. Let’s take a look at the high-level view from the perspective of an Alteryx user and then dive into some of the technical details powering our new one-click solution.

 

For users familiar with the Assisted Modeling experience, you’ll first notice a new step in the modeling process that asks if you would like to enter the classic step-by-step guide for constructing your model or if you would like to enter our new automatic mode. This will give users the ability to choose their best path forward while constructing models for deployment. If you would like to control your modeling steps manually with your own knowledge or insights, the step-by-step process may still be preferred. Otherwise, you can click the “Automatic” button and kick back while our algorithms do the heavy lifting for you.

 

automodelscreen.PNG

After our algorithms crunch the numbers and find the best model for your business use-case, things will look familiar if you have used our previous Assisted Modeling solution. The leaderboard will display the model with the best accuracy, and the steps of the modeling process have been completed for you with the best permutations. You can still examine statistics related to your model’s performance and add that model to your canvas in the form of the machine learning tools or through the Python Tool. Once you have selected the model and added the tools to your canvas, the workflow is ready to predict on new data. The tools are available on the canvas as well if you wish to further tweak your solution based on further statistical insights or just peek at what our automated algorithms have chosen.

 

leaderboard.PNG

So what is happening under the hood when you kick off automatic modeling? We use state-of-the-art data science and statistics algorithms to determine the best classifications of your data types, features and necessary transformations to yield the highest accuracy and the top performance of your model. We then train multiple types of models and compare different scoring metrics to find the model that will best meet your needs and give you the most confident accuracy. We optimize the speed at which we train and compare these models and quickly discard algorithm and model permutations that may not have performed as efficiently in these circumstances. You can trust that the optimum choices have been made for you by modifying the tool properties on the canvas or even by running a separate model through the step-by-step modeling feature to ensure that the automated modeling process has provided you the best parameters and choices for your data.

 

The integration melding the step-by-step assisted mode with the automated step for generating your model makes it seamless to continue your predictive modeling journey without the added hassle or complexity of learning new systems or tools. We are confident that our Alteryx users will enjoy this new functionality and that their journey through the predictive landscape will be much more streamlined as they continue to explore, understand, and infer the current and future state of their data.

 

Closing

Josh and team have done an amazing job of integrating this functionality into the Intelligence Suite to enhance your experience, and as the product leader in this space, I’m constantly driven by seeing the business altering insights our customers enable via our data science innovations in Alteryx.  I look forward to seeing how automating machine learning helps you realize value from your data with even greater velocity than ever before!

 

Download the Intelligence Suite starter kit today to explore pre-built templates with sample data, workflows, and use cases. Contact your account representative to purchase a license of the Alteryx Intelligence Suite, available in Designer in 2020.2 and later so you can start creating and democratizing data science across your organization. If you don't know your rep, or have questions about how to get started, head over to our Support Contact Portal and select Request Access to Intelligence Suite so we can get you set up and ready to solve using code-free advanced analytics in Alteryx.

Alexander Sadovsky
Senior Director of Product

Alex is focused on guiding product teams charged with defining and executing the strategy for Alteryx’s advanced analytics and data science products, along with creating SDKs for the Alteryx end to end platform.

Alex is focused on guiding product teams charged with defining and executing the strategy for Alteryx’s advanced analytics and data science products, along with creating SDKs for the Alteryx end to end platform.