Today organizations are looking into new ways to apply AI to solve unique business problems—from projecting sales to complex manufacturing development—by adding ML models into the DNA of each business function. The main concern of organizations is how to move fast from experimentation to scaling AI without sacrificing trust and transparency.
In this release, DataRobot is excited to announce that Time Series Clustering is now available for SaaS users. In addition, DataRobot also focused on improving model observability with large-scale monitoring with Python, data drift monitoring over time, prediction processing stats, and more.
Also this month DataRobot Dedicated Managed AI Cloud is available for public preview. With this model, AI Cloud is deployed for each customer in a dedicated and separate VPC. By eliminating implementation time and resources, organizations can more quickly apply data engineering, machine learning, decision intelligence, and ML Ops capabilities.
Large-scale Monitoring for Python
Quickly aggregate raw features and predictions into monitoring stats using the Python Library to keep an eye on model performance over time. DataRobot users who currently use the DataRobot-MLOps Python library to report prediction metrics can now also aggregate raw feature and target data using the MLOps-aggregation library. Improve productivity by summarizing necessary monitoring stats—obtain more information faster and with fewer steps.
Drift Over Time
Changes in customer behavior and preferences could be fast and drastic. Data we use for more models can change overnight. You need to react quickly to changes that impact your production AI. With the new Data Drift Over Time View, you have more systematic insights to identify problems and patterns over time. For example, you can drill into a feature that has drifted and see how its drift has changed over time and how its drift relates to other features in the model. A more comprehensive view will help you identify what causes the problem, navigate through the change much faster, and proactively address issues with your model.
Feature Discovery Support with Feature Cache Enabled in No-Code AI Apps
DataRobot No-Code AI Apps now support Feature Discovery projects with feature cache enabled. Take newly created Automated Feature Discovery enabled projects that are built off multiple datasets and build simulations, find insights in a single view, run optimizations, or create what-if scenarios in a seamless, end-to-end flow, entirely within the DataRobot platform.
As seen in the image below, users can easily identify whether features stem from their primary or secondary dataset. Additionally, feature cache is a preview feature that pre-calculates the feature discovery features, and makes them available at prediction time.
Time Series What-if Scenarios
The Time Series What-If App is a No-Code AI App built specifically for Time Series projects. While some users may be familiar with the ability to build Time Series Predictor projects within No Code Apps already, Times Series What-If support expands DataRobot capabilities, allowing you to simulate different scenarios across a specific time window to optimize output. By adjusting known-in-advance features set at the beginning of project creation, you can check the sales impact of running certain marketing campaigns or staffing a certain number of employees for your retail stores over the next seven days. You can edit multiple scenarios simultaneously, resulting in significant time reduction compared to running each scenario sequentially.
Time Series Clustering
The Time Series Clustering Capability—an out of the box solution unique to DataRobot—enables users to easily identify and group similar series across a multi-series dataset. Users previously manually ran a time series clustering technique outside the platform and then used the cluster assigned as a segmenting feature within a time series segmented modeling workflow. Now, this process is seamless and entirely contained within the DataRobot platform. For example, if you are predicting sales of shoes across stores in all of North America, then DataRobot AI Cloud platform will automatically group all stores in San Francisco and Cleveland into one cluster because it detected identical sales profiles for these two locations. Users need not be familiar with advanced concepts like Dynamic Time Warp or code savvy to use the new clustering capability.
DataRobot AI Cloud – September 2022 Release Full Feature list
For the full details of features included in the DataRobot AI Cloud September 2022 Release, visit the DataRobot Documentation Release Center.